Authors: Scott Reed Konrad Żołna Emilio Parisotto Sergio Gómez Colmenarejo Alexander Novikov Gabriel Barth-Maron Mai Giménez Yury Sulsky Jackie Kay Jost Tobias Springenberg Tom Eccles Jake Bruce Ali Razavi Ashley Edwards Nicolas Heess Yutian Chen Raia Hadsell Oriol Vinyals Mahyar Bordbar Nando de Freitas Authors: Scott Reed Konrad Żołna Emilio Parisotto Sergio Gómez Colmenarejo Alexander Novikov Gabriel Barth-Maron Mai Giménez Yury Sulsky Jackie Kay Jost Tobias Springenberg Tom Eccles Jake Bruce Ali Razavi Ashley Edwards Nicolas Heess Yutian Chen Raia Hadsell Oriol Vinyals Mahyar Bordbar Nando de Freitas Scott Reed Konrad Żołna Emilio Parisotto Sergio Gómez Colmenarejo Alexander Novikov Gabriel Barth-Maron Mai Giménez Yury Sulsky Jackie Kay Jost Tobias Springenberg Tom Eccles Jake Bruce Ali Razavi Ashley Edwards Nicolas Heess Yutian Chen Raia Hadsell Oriol Vinyals Mahyar Bordbar Nando de Freitas Abstract Inspired by progress in large-scale language modeling, we apply a similar approach towards building a single generalist agent beyond the realm of text outputs. The agent, which we refer to as Gato, works as a multi-modal, multi-task, multi-embodiment generalist policy. The same network with the same weights can play Atari, caption images, chat, stack blocks with a real robot arm and much more, deciding based on its context whether to output text, joint torques, button presses, or other tokens. In this report we describe the model and the data, and document the current capabilities of Gato. 1 Introduction There are significant benefits to using a single neural sequence model across all tasks. It reduces the need for hand crafting policy models with appropriate inductive biases for each domain. It increases the amount and diversity of training data since the sequence model can ingest any data that can be serialized into a flat sequence. Furthermore, its performance continues to improve even at the frontier of data, compute and model scale (Kaplan et al., 2020; Hoffmann et al., 2022). Historically, generic models that are better at leveraging computation have also tended to overtake more specialized domain-specific approaches (Sutton, 2019), eventually. (Kaplan et al., 2020; Hoffmann et al., 2022). (Sutton, 2019), In this paper, we describe the current iteration of a general-purpose agent which we call Gato, instantiated as a single, large, transformer sequence model. With a single set of weights, Gato can engage in dialogue, caption images, stack blocks with a real robot arm, outperform humans at playing Atari games, navigate in simulated 3D environments, follow instructions, and more. While no agent can be expected to excel in all imaginable control tasks, especially those far outside of its training distribution, we here test the hypothesis that training an agent which is generally capable on a large number of tasks is possible; and that this general agent can be adapted with little extra data to succeed at an even larger number of tasks. We hypothesize that such an agent can be obtained through scaling data, compute and model parameters, continually broadening the training distribution while maintaining performance, towards covering any task, behavior and embodiment of interest. In this setting, natural lan-guage can act as a common grounding across otherwise incompatible embodiments, unlocking combinatorial generalization to new behaviors. large number We focus our training at the operating point of model scale that allows real-time control of real-world robots, currently around 1.2B parameters in the case of Gato. As hardware and model architectures improve, this operating point will naturally increase the feasible model size, pushing generalist models higher up the scaling law curve. For simplicity Gato was trained offline in a purely supervised manner; however, in principle, there is no reason it could not also be trained with either offline or online reinforcement learning (RL). 2 Model The guiding design principle of Gato is to train on the widest variety of relevant data possible, including diverse modalities such as images, text, proprioception, joint torques, button presses, and other discrete and continuous observations and actions. To enable processing this multi-modal data, we serialize all data into a flat sequence of tokens. In this representation, Gato can be trained and sampled from akin to a standard large-scale language model. During deployment, sampled tokens are assembled into dialogue responses, captions, button presses, or other actions based on the context. In the following subsections, we describe Gato’s tokenization, network architecture, loss function, and deployment. 2.1 Tokenization There are infinite possible ways to transform data into tokens, including directly using the raw underlying byte stream. Below we report the tokenization scheme we found to produce the best results for Gato at the current scale using contemporary hardware and model architectures. Text is encoded via SentencePiece (Kudo & Richardson, 2018) with 32000 subwords into the integer range [0, 32000). Images are first transformed into sequences of non-overlapping 16 16 patches in raster order, as done in ViT (Dosovitskiy et al., 2020). Each pixel in the image __p__atches is then normalized between [−1*,* 1] and divided by the square-root of the patch size (i.e. √16 = 4). Discrete values, e.g. Atari button presses, are flattened into sequences of integers in row-major order. The tokenized result is a sequence of integers within the range of [0*,* 1024). Continuous values, e.g. proprioceptive inputs or joint torques, are first flattened into sequences of floating point values in row-major order. The values are mu-law encoded to the range [ 1*,* 1] if not already there (see Figure 14 for details), then discretized to 1024 uniform bins. The discrete integers are then shifted to the range of [32000*,* 33024). Text is encoded via SentencePiece (Kudo & Richardson, 2018) with 32000 subwords into the integer range [0, 32000). (Kudo & Richardson, 2018) Images are first transformed into sequences of non-overlapping 16 16 patches in raster order, as done in ViT (Dosovitskiy et al., 2020). Each pixel in the image __p__atches is then normalized between [−1*,* 1] and divided by the square-root of the patch size (i.e. √16 = 4). (Dosovitskiy et al., 2020). − √ Discrete values, e.g. Atari button presses, are flattened into sequences of integers in row-major order. The tokenized result is a sequence of integers within the range of [0*,* 1024). Continuous values, e.g. proprioceptive inputs or joint torques, are first flattened into sequences of floating point values in row-major order. The values are mu-law encoded to the range [ 1*,* 1] if not already there (see Figure 14 for details), then discretized to 1024 uniform bins. The discrete integers are then shifted to the range of [32000*,* 33024). 14 After converting data into tokens, we use the following canonical sequence ordering. Text tokens in the same order as the raw input text. Image patch tokens in raster order. Tensors in row-major order. Nested structures in lexicographical order by key. Agent timesteps as observation tokens followed by a separator, then action tokens. Agent episodes as timesteps in time order. Text tokens in the same order as the raw input text. Image patch tokens in raster order. Tensors in row-major order. Nested structures in lexicographical order by key. Agent timesteps as observation tokens followed by a separator, then action tokens. Agent episodes as timesteps in time order. Further details on tokenizing agent data are presented in the supplementary material (Section B). B). 2.2 Embedding input tokens and setting output targets After tokenization and sequencing, we apply a parameterized embedding function *f* ( ; *θe*) to each token (i.e. it is applied to both observations and actions) to produce the final model input. To enable efficient learning from our multi-modal input sequence *s*1:*L* the embedding function performs different operations depending on the modality the token stems from: • Tokens belonging to text, discrete- or continuous-valued observations or actions for any time-step are embedded via a lookup table into a learned vector embedding space. Learnable position encodings are added for all tokens based on their local token position within their corresponding time-step. • Tokens belonging to image patches for any time-step are embedded using a single ResNet (He et al., 2016a) block to obtain a vector per patch. For image patch token embeddings, we also add a learnable within-image position encoding vector. (He et al., 2016a) We refer to appendix Section C.3 for full details on the embedding function. C.3 As we model the data autoregressively, each token is potentially also a target label given the previous tokens. Text tokens, discrete and continuous values, and actions can be directly set as targets after tokenization. Image tokens and agent nontextual observations are not currently predicted in Gato, although that may be an interesting direction for future work. Targets for these non-predicted tokens are set to an unused value and their contribution to the loss is masked out. 2.3 Training Given a sequence of tokens s1:L and parameters θ, we model the data using the chain rule of probability: s L θ Let b index a training batch of sequences . We define a masking function *m* such that *m*(*b, l*) = 1 if the token at index *l* is either from text or from the logged action of an agent, and 0 otherwise. The training loss for a batch *B* can then be written as b As described above, Gato’s network architecture has two main components: the parameterized embedding function which transforms tokens to token embeddings, and the sequence model which outputs a distribution over the next discrete token. While any general sequence model can work for next token prediction, we chose a transformer (Vaswani et al., 2017) for simplicity and scalability. Gato uses a 1.2B parameter decoder-only transformer with 24 layers, an embedding size of 2048, and a post-attention feedforward hidden size of 8196 (more details in Section C.1). aswani et al., 2017) C.1). Because distinct tasks within a domain can share identical embodiments, observation formats and action specifications, the model sometimes needs further context to disambiguate tasks. Rather than providing e.g. one-hot task identifiers, we instead take inspiration from (Sanh et al., 2022; Wei et al., 2021; Brown et al., 2020) and use prompt conditioning. During training, for 25% of the sequences in each batch, a prompt sequence is prepended, coming from an episode generated by the same source agent on the same task. Half of the prompt sequences are from the end of the episode, acting as a form of goal conditioning for many domains; and the other half are uniformly sampled from the episode. During evaluation, the agent can be prompted using a successful demonstration of the desired task, which we do by default in all control results that we present here. (Sanh et al., 2022; Wei et al., 2021; Brown et al., 2020) Training of the model is performed on a 16x16 TPU v3 slice for 1M steps with batch size 512 and token sequence length L = 1024, which takes about 4 days. Architecture details can be found in Section C. Because agent episodes and documents can easily contain many more tokens than fit into context, we randomly sample subsequences of L tokens from the available episodes. Each batch mixes subsequences approximately uniformly over domains (e.g. Atari, MassiveWeb, etc.), with some manual upweighting of larger and higher quality datasets (see Table 1 in Section 3 for details). L C. L 1 3 2.4 Deployment Deploying Gato as a policy is illustrated in Figure 3. First a prompt, such as a demonstration, is tokenized, forming the initial sequence. By default, we take the first 1024 tokens of the demonstration. Next the environment yields the first observation which is tokenized and appended to the sequence. Gato samples the action vector autoregressively one token at a time. Once all tokens comprising the action vector have been sampled (determined by the action specification of the environment), the action is decoded by inverting the tokenization procedure described in Section 2.1. This action is sent to the environment which steps and yields a new observation. The procedure repeats. The model always sees all previous observations and actions in its context window of 1024 tokens. We found it beneficial to use transformer XL memory during deployment, although it was not used during training (Dai et al., 2019). 3. 2.1. (Dai et al., 2019). 3 Datasets Gato is trained on a large number of datasets comprising agent experience in both simulated and real world environments, as well as a variety of natural language and image datasets. The datasets we use and their attributes are listed in Table 1. The approximate number of tokens per control dataset is computed assuming the tokenization mechanism described in Section 2.1. 1. 2.1. 3.1 Simulated control tasks Our control tasks consist of datasets generated by specialist SoTA or near-SoTA reinforcement learning agents trained on a variety of different environments. For each environment we record a subset of the experience the agent generates (states, actions, and rewards) while it is training. The simulated environments include Meta-World (Yu et al., 2020) introduced to benchmark meta-reinforcement learning and multi-task learning, Sokoban (Racanière et al., 2017) proposed as a planning problem, BabyAI (Chevalier-Boisvert et al., 2018) for language instruction following in grid-worlds, the DM Control Suite (Tunyasuvunakool et al., 2020) for continuous control, as well as DM Lab (Beattie et al., 2016) designed to teach agents navigation and 3D vision from raw pixels with an egocentric viewpoint. We also use the Arcade Learning Environment (Bellemare et al., 2013) with classic Atari games (we use two sets of games that we call ALE Atari and ALE Atari Extended, see Section F.1 for details). u et al., 2020) (Racanière et al., 2017) (Chevalier-Boisvert et al., 2018) unyasuvunakool et al., 2020) (Beattie et al., 2016) (Bellemare et al., 2013) F.1 We as well include the Procgen Benchmark (Cobbe et al., 2020) and Modular RL (Huang et al., 2020). We also include four tasks using a simulated Kinova Jaco arm from DM Manipulation Playground, as introduced in Zolna et al. (2020). Section F includes a more in-depth description of these control tasks, along with what RL agent was used to generate the data. (Cobbe et al., 2020) (Huang et al., 2020). Zolna et al. (2020). F We found it effective to train on a filtered set of episodes with returns at least 80% of the expert return for the task. The expert return measures the maximum sustained performance that the expert agent can achieve. We define it as the maximum over the set of all windowed average returns calculated over all the collected episodes for a task: where N it the total number of collected episodes for the task, W is the window size, and Ri is the total return for episode i. To obtain accurate estimates, in practice, we set W to be 10% of the total data amount or a minimum of 1000 episodes (i.e. W = min(1000*,* 0*.*1 × N )). N W Ri i W W × N 3.2 Vision and language Gato is trained on MassiveText (Rae et al., 2021), a collection of large English-language text datasets from multiple sources: web pages, books, news articles, and code. (Rae et al., 2021), We also included several vision-language datasets in Gato’s training. ALIGN (Jia et al., 2021) consists of 1.8B images and their alternative text (alt-text) annotations. LTIP (Long Text & Image Pairs), consists of 312 million images with captions (Alayrac et al., 2022). Conceptual captions (Sharma et al., 2018) and COCO captions (Chen et al., 2015) are captioning datasets with 3.3M and 120k image-text pairs respectively. The MultiModal MassiveWeb (M3W) dataset (Alayrac et al., 2022) includes 43M webpages where both text and images were extracted. We also included visual question-answering datasets. In particular OKVQA (Marino et al., 2019) and VQAv2 (Antol et al., 2015) with 9K and 443K triplets of images, questions, and answers. To form a training episode from these, we sample five (image, text) pairs, tokenize them, concatenate, and then pad or randomly crop to the required training sequence length. (Jia et al., 2021) (Alayrac et al. 2022). (Sharma et al., 2018) (Chen et al. 2015) (Alayrac et al 2022) (Marino et al., 2019) (Antol et al., 2015) 3.3 Robotics - RGB Stacking Benchmark (real and sim) As a testbed for taking physical actions in the real world, we chose the robotic block stacking environment introduced by [Lee et al.](#_bookmark89) [(2021).](#_bookmark89) The environment consists of a Sawyer robot arm with 3-DoF cartesian velocity control, an additional DoF for velocity, and a discrete gripper action. The robot’s workspace contains three plastic blocks colored red, green and blue with varying shapes. The available observations include two 128 128 camera images, robot arm and gripper joint angles as well as the robot’s end-effector pose. Notably, ground truth state information for the three objects in the basket is not observed by the agent. Episodes have a fixed length of 400 timesteps at 20 Hz for a total of 20 seconds, and at the end of an episode block positions are randomly re-positioned within the workspace. The robot in action is shown in Figure [4.](#_bookmark8) There are two challenges in this benchmark: *Skill Mastery* (where the agent is provided data from the 5 test object triplets it is later tested on) and *Skill Generalization* (where data can only be obtained from a set of training objects that excludes the 5 test sets). We used several sources of training data for these tasks. In Skill Generalization, for both simulation and real, we use data collected by the best generalist sim2real agent from Lee et al. (2021). We collected data only when interacting with the designated RGB-stacking training objects (this amounts to a total of 387k successful trajectories in simulation and 15k trajectories in real). For Skill Mastery we used data from the best per group experts from Lee et al. (2021) in simulation and from the best sim2real policy on the real robot (amounting to 219k trajectories in total). Note that this data is only included for specific Skill Mastery experiments in Section 5.4. Lee et al. (2021). training objects Lee et al. (2021) 5.4. 4 Capabilities of the generalist agent In this section, we summarize the performance of Gato when trained on the above described data. That is, all results across all tasks are derived from a single pretrained model with a single set of weights. Results with fine-tuning will be presented in Section 5. 5. 4.1 Simulated control tasks Figure 5 shows the number of distinct control tasks for which Gato performs above a given score threshold, relative to expert performance demonstrated in Gato’s training data. 5 We report performance as a percentage, where 100% corresponds to the per-task expert and 0% to a random policy. For each simulated control task we trained our model on, we roll out the Gato policy on the corresponding environment 50 times and average the defined scores. As shown in Figure 5, Gato performs over 450 out of 604 tasks at over a 50% expert score threshold. 5, In ALE Atari(Bellemare et al., 2013) Gato achieves the average human (or better) scores for 23 Atari games1, achieving over twice human score for 11 games. While the single-task online RL agents which generated the data still outperform Gato, this may be overcome by adding capacity or using offline RL training rather than purely supervised (see Section 5.5 where we present a specialist single domain ALE Atari agent achieving better than human scores for 44 games). (Bellemare et al., 2013) 1 5.5 On BabyAI (Chevalier-Boisvert et al., 2018) Gato achieves over 80% of expert score for nearly all levels2. For the most difficult task, called BossLevel, Gato scores 75%. The two other published baselines we could find, BabyAI 1.0 and BabyAI 1.1 (Hui et al., 2020), scored 77% and 90%, respectively, having trained on this single task alone using a million demonstrations. (Chevalier-Boisvert et al., 2018) 2 (Hui et al. 2020), On Meta-World (Yu et al., 2020) Gato achieves more than 50% for all 44 out of 45 tasks that we trained on, over 80% for 35 tasks, and over 90% for 3 tasks. On canonical DM Control Suite (Tassa et al., 2018), Gato achieves better than 50% of the expert score on 21 out of 30 tasks from state, and more than 80% for 18 tasks. u et al., 2020) assa et al., 2018), 4.2 Robotics First person teleoperation enables the collection of expert demonstrations. However, such demonstrations are slow and costly to collect. Data-efficient behavior cloning methods are therefore desirable for training a generalist robot manipulator and offline pretraining is thus a well-motivated area of research. To that end, we evaluated Gato on the established RGB Stacking benchmark for robotics. Skill Generalization Performance The Skill Generalization challenge from the RGB Stacking robotics benchmark tests the agent’s ability to stack objects of previously unseen shapes. The agent is trained on a dataset consisting of episodes of the robot stacking objects with a variety of different shapes. Five triplets of object shapes are, however, not included in the training data and serve as test triplets. We evaluated the trained generalist for 200 episodes per test triplet on the real robot. Table 2 shows that our generalist agent’s success rate on each test triplet is comparable to the single task BC-IMP (filtered BC) baseline in Lee et al. (2021). 2 Lee et al. (2021). 4.3 Text samples The model demonstrates rudimentary dialogue and image captioning capabilities. Figure 6 contains a rep-resentative sample of Gato’s image captioning performance. Figure 7 shows some hand-picked examples of plain text dialogue exchange. 6 7 5 Analysis 5.1 Scaling Laws Analysis In Figure 8, we analyze the aggregate in-distribution performance of the pretrained model as a function of the number of parameters in order to get insight into how performance could improve with increased model capacity. We evaluated 3 different model sizes (measured in parameter count): a 79M model, a 364M model, and a 1.18B model (Gato). We refer to Section C for details on the three model architectures. 8, C Here, for all three model sizes we plot the normalized return as training progresses. To get this single value, for each task we calculate the performance of the model as a percentage of expert score (the same as done in Section 4.1). Then for each domain listed in Table 1 we average the percentage scores across all tasks for that domain. Finally, we mean-aggregate the percentage scores across all domains. We can see that for an equivalent token count, there is a significant performance improvement with increased scale. 4. 1 5.2 Out of distribution tasks In this section we want to answer the following question: Can our agent be used to solve a completely new task efficiently? For this reason, we held-out all data for four tasks from our pre-training set: cartpole.swingup (DM Control Suite domain), assembly-v2 (Meta-World domain), order_of_apples_forage_simple (DM Lab domain), and boxing (ALE Atari domain). These four tasks will serve as testbeds for evaluating the out-of-distribution capabilities of Gato. Can our agent be used to solve a completely new task efficiently? Ideally, the agent could potentially learn to adapt to a new task via conditioning on a prompt including demonstrations of desired behaviour. However, due to accelerator memory constraints and the extremely long sequence lengths of tokenized demonstrations, the maximum context length possible does not allow the agent to attend over an informative-enough context. Therefore, to adapt the agent to new tasks or behaviours, we choose to fine-tune the agent’s parameters on a limited number of demonstrations of a single task, and then evaluate the fine-tuned model’s performance in the environment. Fine-tuning is very similar to pretraining with minor changes, such as different learning rate schedule; see Section E for details. E We want to measure how choice of data used during pretraining influences post-fine-tuning performance. To this end, we compare Gato (trained on all data) to variants trained on ablated datasets: all data 1. A model pretrained only on data from the same domain as the task to be fine-tuned on, same domain only data. same domain only data 2. A model pretrained only on non-control data, no control data. no control data 3. A model fine-tuned from scratch, i.e. no pretraining at all, scratch. scratch Considering as all these experiments require training a new model from scratch and then also fine-tuning, we present results using the less compute-intensive 364M parameter architecture described in Section 5.1. Results are shown in Figure 9. 5.1. 9. Fine-tuning performance on both cartpole.swingup and assembly-v2 tasks, both of which do not require image processing, present similar trends. Pretraining on all the datasets yields the best results, followed by pretraining on the same domain only. This difference is smaller for assembly-v2 but consistent for all few shot datasets. For these non-image-based environments, we see either no benefit (cartpole.swingup) or even negative transfer (assembly-v2) when pretraining on no control datasets, which only contain images and text data. no control Results for DM Lab order_of_apples_forage_simple are slightly different. Pretraining on DM Lab data only is already enough to approach the maximum reward of 19 and hence there is no observable benefit of adding data from different environments. What is different when compared to previously analysed no-vision environments is that pretraining on no control data helps, which can be possibly explained by the fact that agents in the DM Lab environment are fed images which, despite being simulated, are natural looking. Therefore, transfer from image captioning or visual grounded question answering tasks is possible. no control We were not able to observe any benefit from pretraining on boxing. The randomly initialized model seems to work better than any of the pretrained variants considered. We hypothesise that this is caused by the game’s input images being visually very distinct from the other data, suggesting transfer is difficult. We discuss this Atari challenge further in our related work section. 5.3 Fine-tuning on Robotic Stacking Tasks Section 4.2 demonstrates that the base Gato capable of a diverse array of tasks can perform competitively on the RGB Stacking Skill Generalization benchmark. In this section, we would like to answer the following question: *How does our agent improve on robotics tasks when allowed to fine-tune similarly to how we fine-tune on new tasks in Section 5.2? *We consider different model sizes and analyse the impact of pretraining datasets on the Skill Generalization benchmark, as well as a novel out of distribution task. Further analysis of fine-tuning with dataset ablations is in Appendix I. 4.2 5.2? I. Skill Generalization First, we would like to show that fine-tuning on object-specific data, similarly to what was done by Lee et al. (2022), is beneficial. Therefore, we fine-tuned Gato separately on five subsets of demonstrations from the test dataset. Each subset was obtained by random partitioning of a test dataset consisting of demonstrations gathered by a generalist sim-to-real agent stacking real test objects. We consider this setting, which is comparable to the fine-tuning baselines on RGB stacking tasks from (Lee et al., 2022); and use the 5k dataset that their behavior cloning 5k results are obtained with. To best match their experiments, we change our return filtering scheme during training: instead of using only successful stacks, we condition on the normalized return of the episode. Lee et al. (2022), test (Lee et al., 2022); Figure 10 compares the success rate of Gato across different fine-tuning data regimes to the sim-to-real expert and a Critic-Regularized Regression (CRR) (Wang et al., 2020) agent trained on 35k episodes of all test triplets. Gato, in both reality and simulation (red curves on the left and right figure, respectively), recovers the expert’s performance with only 10 episodes, and peaks at 100 or 1000 episodes of fine-tuning data, where it exceeds the expert. After this point (at 5000), performance degrades slightly but does not drop far below the expert’s performance. 10 (Wang et al., 2020) Fine-tuning and Model Size To better understand the benefit of large models for few-shot adaptation in robotics domains, we conducted an ablation on model parameter size. This section focuses on in-simulation evaluation. Figure 10 compares the full 1.18B parameter Gato with the smaller 364M and 79M parameter variants for varying amounts of fine-tuning data. Although the 364M model overfits on one episode, causing performance to drop, there is a clear trend towards better adaptation with fewer episodes as the number of parameters is scaled up. The 79M model performs clearly worse than its bigger counterparts. The results suggest that the model’s greater capacity allows the model to use representations learned from the diverse training data at test time. 10 Adaptation to Perceptual Variations While the Skill Generalization task is an effective benchmark for motor Skill Generalization to shape varia-tions, it does not test the agent’s ability to adapt to perceptual variations and permutations in the objective specification. To further evaluate Gato’s generalization capabilities, we devised a new task in the RGB stacking benchmark where the goal is to stack the blue object on the green object, for test triplet 1 (see Figure 11). First, we used a 3D mouse to collect 500 demonstrations of this task on the real robot, for a total of 2 hours and 45 minutes of demonstration data, and fine-tuned Gato on these episodes. Notably, all of the simulated and real robotics data in the pretraining set shows the robot successfully stacking the red object on the blue object, and the data does not include the object shapes in the test set. We found that additionally adding simulated demonstrations of the stack blue on green task to the fine-tuning dataset improved performance, and 10% was an ideal sampling ratio for this data. 11). We achieved a final 60% success rate after evaluating fine-tuned Gato on the real robot, while a BC baseline trained from scratch on the blue-on-green data achieved only 0.5% success (1/200 episodes). Qualitatively, the BC baseline would consistently move towards the blue object and occasionally pick it up and place it on top of the green object, but a full, stable stack was almost never achieved. 5.4 Robotics: Skill Mastery Similarly to the Skill Generalization challenge discussed in Section 4.2, the Skill Mastery challenge consists in training a robotic arm to stack blocks of different shapes. However, the Skill Mastery allows the agent to train on data involving the object shapes used for evaluation, i.e. the test set in Skill Generalization becomes a part of the Skill Mastery training set. Thus, this challenge serves to measure Gato’s performance on in-distribution tasks (possibly with initial conditions not seen in the training demonstrations). Our Skill Mastery results use an earlier version of the Gato architecture described in Appendix H, with no fine-tuning. 4.2, test training H, Table 3 compares the group-wise success percentage and the average success across object groups for Gato and the established BC-IMP baseline. Gato exceeds or closely matches BC-IMP’s performance on all but one training triplet. 3 5.5 Specialist single-domain multi-task agents In this section we show results obtained with two specialist (rather than generalist) agents. Both of them were trained on data from a single domain only and rolled out 500 times for each training task without any per-task fine-tuning. Meta-World The first agent uses the smallest architecture introduced in Section 5.1, i.e. 79M parameters, and is trained on all 50 Meta-World tasks. While Gato has access to the state of the MuJoCo physics engine and unlimited task seeds, the agent presented here has no access to any extra features or tasks and uses the canonical API as in (Yu et al., 2020). This experiment is to show that the architecture proposed in our paper can be used to obtain state-of-the-art agents also at small scale. The training procedure was to train single-task MPO (Abdolmaleki et al., 2018) experts on each of the MT-50 tasks individually, recording the trajectories produced while training. This experience is then combined, or distilled, into a single agent, which achieves 96.6% success rate averaged over all 50 tasks. To the best of our knowledge this agent is the first one to accomplish nearly 100% average success rate simultaneously (multi-task) for this benchmark. See Table 7 in the supplementary material (Section K) for the full list of tasks and corresponding success rates of our agent. 5.1, u et al., 2020). (Abdolmaleki et al., 2018) 7 K) ALE Atari We also trained a specialist agent on all 51 ALE Atari tasks. As the Atari domain is much more challenging than Meta-World, we used the Gato architecture with 1.18B parameters. The resulting agent performs better than the average human for 44 games (see Section 4.1 for details on our evaluation and scoring). We want to note that the performance of online experts used to generate training data for the other 7 games were also below the average human. Hence, the specialist Atari agent achieved better than human performance for all games where data contained super-human episodes. 4.1 The specialist Atari agent outperforms our generalist agent Gato, which achieved super-human performance on 23 games. It suggests that scaling Gato may result in even better performance. We, however, purposely restricted Gato’s size such that it can be run in real-time on the real robot. 5.6 Attention Analysis We rendered the transformer attention weights over the image observations for various tasks, to gain a qualitative sense of how Gato attends to different regions of the image across tasks (see Figure 12). Further details and visualizations for more tasks can be found in Appendix J. These visualizations clearly show that attention tracks the task-relevant objects and regions. 12). J. 5.7 Embedding Visualization To understand how Gato encodes differently information per task, we visualized per-task embeddings. We analysed 11 tasks. For each task, we randomly sample 100 episodes and tokenize each of them. Then, from each episode we take a subsequence of 128 tokens, compute their embeddings (at layer 12, which is half the total depth of the transformer layers) and average them over the sequence. The averaged embeddings for all tasks are used as input to PCA, which reduces their dimensionality to 50. Then, T-SNE is used to get the final 2D embeddings. Figure 13 shows the final T-SNE embeddings plotted in 2D, colorized by task. Embeddings from the same tasks are clearly clustered together, and task clusters from the same domain and modality are also located close to each other. Even held-out task (cartpole.swingup) is clustered correctly and lays next to another task from DM Control Suite Pixels. 13 6 Related Work The most closely related architectures to that of Gato are Decision Transformers (Chen et al., 2021b; Reid et al., 2022; Zheng et al., 2022; Furuta et al., 2021) and Trajectory Transformer (Janner et al., 2021), which showed the usefulness of highly generic LM-like architectures for a variety of control problems. Gato also uses an LM-like architecture for control, but with design differences chosen to support multi-modality, multi-embodiment, large scale and general purpose deployment. Pix2Seq (Chen et al., 2022) also uses an LM-based architecture for object detection. Perceiver IO (Jaegle et al., 2021) uses a transformer-derived architecture specialized for very long sequences, to model any modality as a sequence of bytes. This and similar architectures could be used to expand the range of modalities supported by future generalist models. (Chen et al., 2021b; Reid et al., 2022; Zheng et al., 2022; Furuta et al. 2021) (Janner et al., 2021), (Chen et al., 2022) (Jaegle et al 2021) Gato was inspired by works such as GPT-3 (Brown et al., 2020) and Gopher (Rae et al., 2021), pushing the limits of generalist language models; and more recently the Flamingo (Alayrac et al., 2022) generalist visual language model. Chowdhery et al. (2022) developed the 540B parameter Pathways Language Model (PalM) explicitly as a generalist few-shot learner for hundreds of text tasks. (Brown et al., 2020) (Rae et al., 2021), (Alayrac et al., 2022) Chowdhery et al. (2022) Future work should consider how to unify these text capabilities into one fully generalist agent that can also act in real time in the real world, in diverse environments and embodiments. Gato also takes inspiration from recent works on multi-embodiment continuous control. Huang et al. (2020) used message passing graph networks to build a single locomotor controller for many simulated 2D walker variants. Kurin et al. (2020) showed that transformers can outperform graph based approaches for incom-patible (i.e. varying embodiment) control, despite not encoding any morphological inductive biases. Devin et al. (2017) learn a modular policy for multi-task and multi-robot transfer in simulated 2D manipulation environments. Chen et al. (2018) train a universal policy conditioned on a vector representation of robot hardware, showing successful transfer both to simulated held out robot arms, and to a real world sawyer robot arm. Huang et al. (2020) Kurin et al. (2020) Devin et al. (2017) Chen et al. (2018) A variety of earlier generalist models have been developed that, like Gato, operate across highly distinct domains and modalities. NPI (Reed & De Freitas, 2016) trained a single LSTM (Hochreiter & Schmidhuber, 1997) to execute diverse programs such as sorting an array and adding two numbers, such that the network is able to generalize to larger problem instances than those seen during training. Kaiser et al. (2017) developed the MultiModel that trains jointly on 8 distinct speech, image and text processing tasks including classifica-tion, image captioning and translation. Modality-specific encoders were used to process text, images, audio and categorical data, while the rest of the network parameters are shared across tasks. Schmidhuber (2018) proposed “one big net for everything”, describing a method for the incremental training of an increasingly general problem solver. Keskar et al. (2019) proposed controllable multi-task language models that can be directed according to language domain, subdomain, entities, relationships between entities, dates, and task-specific behavior. (Reed & De Freitas, 2016) (Hochreiter & Schmidhuber, 1997) Kaiser et al. (2017) Schmidhuber (2018) one big net for everything Keskar et al. (2019) In this discussion, it is important to distinguish between one single multi-task network architecture versus one single neural network with the same weights for all tasks. Several poplar RL agents achieve good multi-task RL results within single domains such as Atari57 and DMLab (Espeholt et al., 2018; Song et al., 2020; Hessel et al., 2019). However, it is much more common to use the same policy architecture and hyper-parameters across tasks, but the policy parameters are different in each task (Mnih et al., 2015; Tassa et al., 2018). This is also true of state-of-the-art RL methods applied to board games (Schrittwieser et al., 2020). Moreover, this choice has been adopted by off-line RL benchmarks (Gulcehre et al., 2020; Fu et al., 2020) and recent works on large sequence neural networks for control, including decision transformers (Chen et al., 2021b; Reid et al., 2022; Zheng et al., 2022) and the Trajectory Transformer of Janner et al. (2021). In contrast, in this work we learn a single network with the same weights across a diverse set of tasks. (Espeholt et al., 2018; Song et al., 2020; Hessel et al., 2019). (Mnih et al., 2015; Tassa et al., 2018). (Schrittwieser et al., 2020). (Gulcehre et al., 2020; Fu et al., 2020) (Chen et al., 2021b; Reid et al., 2022; Zheng et al., 2022) Janner et al. (2021). Recent position papers advocate for highly generalist models, notably Schmidhuber (2018) proposing one big net for everything, and Bommasani et al. (2021) on foundation models. However, to our knowledge there has not yet been reported a single generalist trained on hundreds of vision, language and control tasks using modern transformer networks at scale. Schmidhuber (2018) Bommasani et al. (2021) “Single-brain”-style models have interesting connections to neuroscience. Mountcastle (1978) famously stated that “the processing function of neocortical modules is qualitatively similar in all neocortical regions. Put shortly, there is nothing intrinsically motor about the motor cortex, nor sensory about the sensory cortex ”. Mountcastle found that columns of neurons in the cortex behave similarly whether associated with vision, hearing or motor control. This has motivated arguments that we may only need one algorithm or model to build intelligence (Hawkins & Blakeslee, 2004). Mountcastle (1978) the processing function of neocortical modules is qualitatively similar in all neocortical regions. Put shortly, there is nothing intrinsically motor about the motor cortex, nor sensory about the sensory cortex (Hawkins & Blakeslee, 2004). Sensory substitution provides another argument for a single model (Bach-y Rita & Kercel, 2003). For example, it is possible to build tactile visual aids for blind people as follows. The signal captured by a camera can be sent via an electrode array on the tongue to the brain. The visual cortex learns to process and interpret these tactile signals, endowing the person with some form of “vision”. Suggesting that, no matter the type of input signal, the same network can process it to useful effect. (Bach-y Rita & Kercel, 2003). Our work is based on deep autoregressive models, which have a long history and can be found in generative models of text, images, video and audio. Combining autoregressive generation with transformers (Vaswani et al., 2017; Devlin et al., 2018) has been of enormous impact in language modelling (Brown et al., 2020; Rae et al., 2021), protein folding (Jumper et al., 2021), vision-language models (Tsimpoukelli et al., 2021; Wang et al., 2021; Alayrac et al., 2022), code generation (Chen et al., 2021c; Li et al., 2022b), dialogue systems with retrieval capabilities (Nakano et al., 2021; Thoppilan et al., 2022), speech recognition (Pratap et al., 2020), neural machine translation (Johnson et al., 2019) and more (Bommasani et al., 2021). Recently researchers have explored task decomposition and grounding with language models (Huang et al., 2022; Ahn et al., 2022). aswani et al., 2017; Devlin et al., 2018) (Brown et al., 2020; Rae et al., 2021), (Jumper et al., 2021), simpoukelli et al., 2021; Wang et al., 2021; Alayrac et al., 2022), (Chen et al., 2021c; Li et al., 2022b), (Nakano et al., 2021; Thoppilan et al., 2022), (Pratap et al., 2020), (Johnson et al., 2019) (Bommasani et al. 2021). (Huang et al., 2022; Ahn et al., 2022). Li et al. (2022a) construct a control architecture, consisting of a sequence tokenizer, a pretrained language model and a task-specific feed-forward network. They apply it to VirtualHome and BabyAI tasks, and find that the inclusion of the pretrained language model improves generalisation to novel tasks. Similarly, Parisi et al. (2022) demonstrate that vision models pretrained with self-supervised learning, especially crop segmentations and momentum contrast (He et al., 2020), can be effectively incorporated into control policies. Li et al. (2022a) Parisi et al. (2022) (He et al., 2020), As mentioned earlier, transfer in Atari is challenging. Rusu et al. (2016) researched transfer between ran-domly selected Atari games. They found that Atari is a difficult domain for transfer because of pronounced differences in the visuals, controls and strategy among the different games. Further difficulties that arise when applying behaviour cloning to video games like Atari are discussed by Kanervisto et al. (2020). Rusu et al. (2016) Kanervisto et al. (2020). There has been great recent interest in data-driven robotics (Cabi et al., 2019; Chen et al., 2021a). However, Bommasani et al. (2021) note that in robotics “the key stumbling block is collecting the right data. Unlike language and vision data, robotics data is neither plentiful nor representative of a sufficiently diverse array of embodiments, tasks, and environments”. Moreover, every time we update the hardware in a robotics lab, we need to collect new data and retrain. We argue that this is precisely why we need a generalist agent that can adapt to new embodiments and learn new tasks with few data. (Cabi et al., 2019; Chen et al., 2021a). Bommasani et al. (2021) the key stumbling block is collecting the right data. Unlike language and vision data, robotics data is neither plentiful nor representative of a sufficiently diverse array of embodiments, tasks, and environments Generating actions using an autoregressive model can lead to causal “self-delusion” biases when there are confounding variables (Ortega et al., 2021). For example, sampling actions can condition the model to solve the wrong task when multiple tasks share similar observation and actions specifications. As explained in Section 2, we use prompt engineering in ambiguous tasks, conditioning our model on a successful demon-stration. This screens off confounding variables, reducing self-delusions. Another solution which we did not explore in this work is to use counterfactual teaching, where we train a model online using instantaneous expert feedback. We leave this for future investigation. (Ortega et al., 2021). 2, 7 Broader Impact Although generalist agents are still only an emerging area of research, their potential impact on society calls for a thorough interdisciplinary analysis of their risks and benefits. For the sake of transparency, we document the intended use cases of Gato in the model card in Appendix A. However, the tools for mitigating harms of generalist agents are relatively underdeveloped, and require further research before these agents are deployed. A. Since our generalist agent can act as a vision-language model, it inherits similar concerns as discussed in (Wei-dinger et al., 2021; Bommasani et al., 2021; Rae et al., 2021; Alayrac et al., 2022). In addition, generalist agents can take actions in the the physical world; posing new challenges that may require novel mitigation strategies. For example, physical embodiment could lead to users anthropomorphizing the agent, leading to misplaced trust in the case of a malfunctioning system, or be exploitable by bad actors. Additionally, while cross-domain knowledge transfer is often a goal in ML research, it could create unexpected and undesired outcomes if certain behaviors (e.g. arcade game fighting) are transferred to the wrong context. The ethics and safety considerations of knowledge transfer may require substantial new research as generalist systems advance. (Wei-dinger et al., 2021; Bommasani et al., 2021; Rae et al., 2021; Alayrac et al., 2022). Technical AGI safety (Bostrom, 2017) may also become more challenging when considering generalist agents that operate in many embodiments. For this reason, preference learning, uncertainty modeling and value alignment (Russell, 2019) are especially important for the design of human-compatible generalist agents. It may be possible to extend some of the value alignment approaches for language (Ouyang et al., 2022; Kenton et al., 2021) to generalist agents. However, even as technical solutions are developed for value alignment, generalist systems could still have negative societal impacts even with the intervention of well-intentioned designers, due to unforeseen circumstances or limited oversight (Amodei et al., 2016). This limitation underscores the need for a careful design and a deployment process that incorporates multiple disciplines and viewpoints. (Bostrom, 2017) ussell, 2019) (Ouyang et al., 2022; Kenton et al., 2021) (Amodei et al., 2016). Understanding how the models process information, and any emergent capabilities, requires significant ex-perimentation. External retrieval (Borgeaud et al., 2021; Menick et al., 2022; Nakano et al., 2021; Thoppilan et al., 2022) has been shown to improve both interpretability and performance, and hence should be consid-ered in future designs of generalist agents. (Borgeaud et al., 2021; Menick et al., 2022; Nakano et al., 2021; Thoppilan et al., 2022) Although still at the proof-of-concept stage, the recent progress in generalist models suggests that safety researchers, ethicists, and most importantly, the general public, should consider their risks and benefits. We are not currently deploying Gato to any users, and so anticipate no immediate societal impact. However, given their potential impact, generalist models should be developed thoughtfully and deployed in a way that promotes the health and vitality of humanity. 8 Limitations and Future work 8.1 RL data collection Gato is a data-driven approach, as it is derived from imitation learning. While natural language or image datasets are relatively easy to obtain from the web, a web-scale dataset for control tasks is not currently available. This may seem at first to be problematic, especially when scaling Gato to a higher number of parameters. That being said, there has already been extensive investigation into this issue. Offline RL aims at leveraging existing control datasets, and its increasing popularity has already resulted in the availability of more diverse and larger datasets. Richer environments and simulations are being built (e.g. Metaverse), and increasing numbers of users already interact with them among thousands of already deployed online games (e.g. there exists a large dataset of Starcraft 2 games). Real-life data has also been already stored for ML research purposes; for example, data for training self-driving cars is acquired from recording human driver data. Finally, while Gato uses data consisting of both observations and corresponding actions, the possibility of using large scale observation-only data to enhance agents has been already studied (Baker et al., 2022). (Baker et al., 2022). Thanks to online video sharing and streaming platforms such as Youtube and Twitch, observation-only datasets are not significantly more difficult to collect than natural language datasets, motivating a future research direction to extend Gato to learn from web data. While the previous paragraph focuses on alleviating drawbacks of data collection from RL agents, it is important to note that this approach presents a different set of tradeoffs compared to scraping web data and can be actually more practical in some situations. Once the simulation is set up and near SOTA agent trained, it can be used to generate massive amounts of high quality data. That is in contrast to the quality of web data which is notorious for its low quality. In short, we believe that acquiring suitable data is another research question on its own, and this is an active area of research with growing momentum and importance. 8.2 Prompt and short context Gato is prompted with an expert demonstration, which aids the agent to output actions corresponding to the given task. This is particularly useful since there is otherwise no task identifier available to the agent (that is in contrast to many multi-task RL settings). Gato infers the relevant task from the observations and actions in the prompt. However, the context length of our agent is limited to 1024 tokens which translates to the agent sometimes attending to only a few environment timesteps in total. This is especially the case for environments with image observations, where depending on the resolution each observation can result in more than one hundred tokens each. Hence for certain environments only a short chunk of a demonstration episode fits in the transformer memory. Due to this limited prompt context, preliminary experiments with different prompt structures resulted in very similar performance. Similarly, early evaluations of the model using prompt-based in-context learning on new environments did not show a significant performance improvement compared to prompt-less evaluation in the same setting. Context-length is therefore a current limitation of our architecture, mainly due to the quadratic scaling of self-attention. Many recently proposed architectures enable a longer context at greater efficiency and these innovations could potentially improve our agent performance. We hope to explore these architectures in future work. 9 Conclusions Transformer sequence models are effective as multi-task multi-embodiment policies, including for real-world text, vision and robotics tasks. They show promise as well in few-shot out-of-distribution task learning. In the future, such models could be used as a default starting point via prompting or fine-tuning to learn new behaviors, rather than training from scratch. Given scaling law trends, the performance across all tasks including dialogue will increase with scale in parameters, data and compute. Better hardware and network architectures will allow training bigger models while maintaining real-time robot control capability. By scaling up and iterating on this same basic approach, we can build a useful general-purpose agent. Acknowledgments We would like to thank Dan Horgan, Manuel Kroiss, Mantas Pajarskas, and Thibault Sottiaux for their help with data storage infrastructure; Jean-Baptiste Lespiau and Fan Yang for help on concurrent evalua-tion; Joel Veness for advising on the model design; Koray Kavukcuoglu for helping inspire the project and facilitating feedback; Tom Erez for advising on the agent design and task selection for continuous control; Igor Babuschkin for helping code the initial prototype; Jack Rae for advising on the transformer language model codebase; Thomas Lampe for building robot infrastructure and advising on real robotics experiments; Boxi Wu for input on ethics and safety considerations; Pedro A. Ortega for advice in regard to causality and self-delusion biases. Author Contributions Scott Reed developed the project concept, wrote the initial prototype, and led the project overall. Konrad Żołna led architecture development for vision and text, built infrastructure for tokenization and prompting, and contributed heavily to overall agent development and evaluation. Scott Reed Konrad Żołna Emilio Parisotto led work on optimizing the transformer architecture, ran the largest number of experi-ments, and analyzed scaling law properties and in-distribution agent performance. Emilio Parisotto Sergio Gómez Colmenarejo was the technical lead, responsible for creating a scalable data loader and evaluator supporting hundreds of tasks at once, and for the initial robot integration with Gato. Sergio Gómez Colmenarejo Alexander Novikov developed the model including the sampler for the initial prototype, carried out ex-periments focusing on robotics, and created visualizations. Alexander Novikov Gabriel Barth-Maron built scalable storage infrastructure to provide Gato with SoTA-level agent expe-rience in Atari and other domains. Gabriel Barth-Maron Mai Giménez conducted large scale agent data collection, built substantial data loading infrastructure, and integrated large scale visual-language datasets into the training of Gato. Mai Giménez Yury Sulsky contributed broadly to the Gato codebase including a bespoke distributed training sequence loader, and led the development of benchmarks for out-of-distribution generalization, and the training of competitive baseline agents. Yury Sulsky Jackie Kay supported physical robotics infrastructure, conducted numerous evaluations and experiments to analyze the generalization properties of Gato, and contemplated broader ethical impact. Jackie Kay Jost Tobias Springenberg guided Gato’s deployment to the physical robot, provided strong existing base-lines for block stacking, and advised on model development and experimental design. Jost Tobias Springenberg Tom Eccles developed the Gato dialogue and image captioning demonstrations, allowing users to easily probe the vision and language capacities of agents in development. Tom Eccles Jake Bruce contributed to agent design as well as control datasets and environments with randomized physics and morphology variations. Jake Bruce Ali Razavi helped in exploring vision architectures. Ali Razavi Ashley Edwards contributed to the first prototype of Gato that worked on Atari, in addition to exploring alternative network architectures and training objectives. Ashley Edwards Nicolas Heess advised on agent design, experiment design and task selection, especially for continuous control applications. Nicolas Heess Yutian Chen advised on model design and experiments, and provided feedback in regular meetings. Yutian Chen Raia Hadsell advised on the design and planning of robotics efforts. Raia Hadsell Oriol Vinyals advised on all aspects of the project, especially model architecture, training strategies and benchmark design. Oriol Vinyals Mahyar Bordbar was the primary project manager; eliciting key goals, tracking progress, facilitating pre-sentations and feedback, and coordinating resource planning. Mahyar Bordbar Nando de Freitas oversaw the project from its inception. Nando de Freitas References Abbas Abdolmaleki, Jost Tobias Springenberg, Yuval Tassa, Remi Munos, Nicolas Heess, and Martin Ried-miller. Maximum a posteriori policy optimisation. Preprint arXiv:1806.06920, 2018. Preprint arXiv:1806.06920 Samira Abnar and Willem Zuidema. Quantifying attention flow in transformers. Preprint arXiv:2005.00928, 2020. Preprint arXiv:2005.00928 Michael Ahn, Anthony Brohan, Noah Brown, Yevgen Chebotar, Omar Cortes, Byron David, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, et al. Do as i can, not as i say: Grounding language in robotic affordances. Preprint arXiv:2204.01691, 2022. Preprint arXiv:2204.01691 Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Mensch, Katie Millican, Malcolm Reynolds, Roman Ring, Eliza Rutherford, Serkan Cabi, Tengda Han, Zhitao Gong, Sina Samangooei, Marianne Monteiro, Jacob Menick, Sebastian Borgeaud, Andy Brock, Aida Nematzadeh, Sahand Sharifzadeh, Mikolaj Binkowski, Ricardo Barreira, Oriol Vinyals, Andrew Zisserman, and Karen Simonyan. Flamingo: a visual language model for few-shot learning. Preprint arXiv:2204.14198, 2022. Preprint arXiv:2204.14198 Dario Amodei, Chris Olah, Jacob Steinhardt, Paul F. Christiano, John Schulman, and Dan Mané. Concrete problems in AI safety. Preprint arXiv:1606.06565, 2016. Preprint arXiv:1606.06565 Stanislaw Antol, Aishwarya Agrawal, Jiasen Lu, Margaret Mitchell, Dhruv Batra, C Lawrence Zitnick, and Devi Parikh. VQA: Visual question answering. In International Conference on Computer Vision, pp. 2425–2433, 2015. International Conference on Computer Vision Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. Preprint arXiv:1607.06450, 2016. Preprint arXiv:1607.06450 Paul Bach-y Rita and Stephen W Kercel. Sensory substitution and the human-machine interface. Trends in cognitive sciences, 7(12):541–546, 2003. Trends in cognitive sciences Bowen Baker, Ilge Akkaya, Peter Zhokhov, Joost Huizinga, Jie Tang, Adrien Ecoffet, Brandon Houghton, Raul Sampedro, and Jeff Clune. Video pretraining (vpt): Learning to act by watching unlabeled online videos. Preprint arXiv::2206.11795, 2022. Preprint arXiv::2206.11795 Gabriel Barth-Maron, Matthew W Hoffman, David Budden, Will Dabney, Dan Horgan, Dhruva Tb, Alistair Muldal, Nicolas Heess, and Timothy Lillicrap. Distributed distributional deterministic policy gradients. Preprint arXiv:1804.08617, 2018. Preprint arXiv:1804.08617 Charles Beattie, Joel Z Leibo, Denis Teplyashin, Tom Ward, Marcus Wainwright, Heinrich Küttler, Andrew Lefrancq, Simon Green, Víctor Valdés, Amir Sadik, et al. DeepMind lab. Preprint arXiv:1612.03801, 2016. Preprint arXiv:1612.03801 Marc G Bellemare, Yavar Naddaf, Joel Veness, and Michael Bowling. The arcade learning environment: An evaluation platform for general agents. Journal of Artificial Intelligence Research, 47:253–279, 2013. Journal of Artificial Intelligence Research Rishi Bommasani, Drew A Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, et al. On the opportunities and risks of foundation models. Preprint arXiv:2108.07258, 2021. Preprint arXiv:2108.07258 Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George van den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, et al. Improving language models by retrieving from trillions of tokens. Preprint arXiv:2112.04426, 2021. Preprint arXiv:2112.04426 Nick Bostrom. Superintelligence. Dunod, 2017. Superintelligence Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. Openai gym. Preprint arXiv:1606.01540, 2016. Preprint arXiv:1606.01540 TB Brown, B Mann, N Ryder, M Subbiah, J Kaplan, P Dhariwal, A Neelakantan, P Shyam, G Sastry, A Askell, et al. Language models are few-shot learners. In Advances in Neural Information Processing Systems, pp. 1877–1901, 2020. Advances in Neural Information Processing Systems Serkan Cabi, Sergio Gómez Colmenarejo, Alexander Novikov, Ksenia Konyushkova, Scott Reed, Rae Jeong, Konrad Zolna, Yusuf Aytar, David Budden, Mel Vecerik, et al. Scaling data-driven robotics with reward sketching and batch reinforcement learning. Preprint arXiv:1909.12200, 2019. Preprint arXiv:1909.12200 Annie S Chen, Suraj Nair, and Chelsea Finn. Learning generalizable robotic reward functions from “in-the-wild" human videos. Preprint arXiv:2103.16817, 2021a. Preprint arXiv:2103.16817 Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Misha Laskin, Pieter Abbeel, Ar-avind Srinivas, and Igor Mordatch. Decision transformer: Reinforcement learning via sequence modeling. Advances in Neural Information Processing Systems, 34, 2021b. Advances in Neural Information Processing Systems Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. Evaluating large language models trained on code. Preprint arXiv:2107.03374, 2021c. Preprint arXiv:2107.03374 Tao Chen, Adithyavairavan Murali, and Abhinav Gupta. Hardware conditioned policies for multi-robot transfer learning. Advances in Neural Information Processing Systems, 31, 2018. Advances in Neural Information Processing Systems Ting Chen, Saurabh Saxena, Lala Li, David J Fleet, and Geoffrey Hinton. Pix2seq: A language modeling framework for object detection. In ICLR, 2022. ICLR Xinlei Chen, Hao Fang, Tsung-Yi Lin, Ramakrishna Vedantam, Saurabh Gupta, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco captions: Data collection and evaluation server. Preprint arXiv:1504.00325, 2015. Preprint arXiv:1504.00325 Maxime Chevalier-Boisvert, Dzmitry Bahdanau, Salem Lahlou, Lucas Willems, Chitwan Saharia, Thien Huu Nguyen, and Yoshua Bengio. BabyAI: A platform to study the sample efficiency of grounded language learning. Preprint arXiv:1810.08272, 2018. Preprint arXiv:1810.08272 Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. PaLM: Scaling language modeling with pathways. Preprint arXiv:2204.02311, 2022. Preprint arXiv:2204.02311 Karl Cobbe, Chris Hesse, Jacob Hilton, and John Schulman. Leveraging procedural generation to benchmark reinforcement learning. In International Conference on Machine Learning, pp. 2048–2056, 2020. International Conference on Machine Learning Zihang Dai, Zhilin Yang, Yiming Yang, Jaime G Carbonell, Quoc Le, and Ruslan Salakhutdinov. Transformer-xl: Attentive language models beyond a fixed-length context. In Annual Meeting of the Association for Computational Linguistics, pp. 2978–2988, 2019. Annual Meeting of the Association for Computational Linguistics Coline Devin, Abhishek Gupta, Trevor Darrell, Pieter Abbeel, and Sergey Levine. Learning modular neural network policies for multi-task and multi-robot transfer. In IEEE International Conference on Robotics & Automation, pp. 2169–2176, 2017. IEEE International Conference on Robotics & Automation Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirec-tional transformers for language understanding. Preprint arXiv:1810.04805, 2018. Preprint arXiv:1810.04805 Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Un-terthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. Preprint arXiv:2010.11929, 2020. Preprint arXiv:2010.11929 Lasse Espeholt, Hubert Soyer, Remi Munos, Karen Simonyan, Vlad Mnih, Tom Ward, Yotam Doron, Vlad Firoiu, Tim Harley, Iain Dunning, et al. Impala: Scalable distributed deep-RL with importance weighted actor-learner architectures. In International Conference on Machine Learning, pp. 1407–1416, 2018. International Conference on Machine Learning Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4RL: Datasets for deep data-driven reinforcement learning. Preprint arXiv:2004.07219, 2020. Preprint arXiv:2004.07219 Hiroki Furuta, Yutaka Matsuo, and Shixiang Shane Gu. Generalized decision transformer for offline hindsight information matching. Preprint arXiv:2111.10364, 2021. Preprint arXiv:2111.10364 Caglar Gulcehre, Ziyu Wang, Alexander Novikov, Thomas Paine, Sergio Gómez, Konrad Zolna, Rishabh Agarwal, Josh S Merel, Daniel J Mankowitz, Cosmin Paduraru, et al. RL unplugged: A suite of benchmarks for offline reinforcement learning. Advances in Neural Information Processing Systems, 33:7248–7259, 2020. Advances in Neural Information Processing Systems Jeff Hawkins and Sandra Blakeslee. On intelligence. Macmillan, 2004. On intelligence Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In IEEE Computer Vision and Pattern Recognition, pp. 770–778, 2016a. IEEE Computer Vision and Pattern Recognition Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Identity mappings in deep residual networks. In European Conference on Computer Vision, pp. 630–645, 2016b. European Conference on Computer Vision Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual representation learning. In IEEE Computer Vision and Pattern Recognition, pp. 9729–9738, 2020. IEEE Computer Vision and Pattern Recognition Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (GELUs). Preprint arXiv:1606.08415, 2016. Matteo Hessel, Hubert Soyer, Lasse Espeholt, Wojciech Czarnecki, Simon Schmitt, and Hado van Hasselt. Preprint arXiv:1606.08415 Multi-task deep reinforcement learning with popart. In AAAI, 2019. AAAI Matteo Hessel, Ivo Danihelka, Fabio Viola, Arthur Guez, Simon Schmitt, Laurent Sifre, Theophane Weber, David Silver, and Hado van Hasselt. Muesli: Combining improvements in policy optimization. Preprint arXiv:2104.06159, 2021. Preprint arXiv:2104.06159 Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural computation, 9(8):1735–1780, 1997. Neural computation Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, et al. Training compute-optimal large language models. Preprint arXiv:2203.15556, 2022. Preprint arXiv:2203.15556 Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Weinberger. Deep networks with stochastic depth. Preprint arXiv:1603.09382, 2016. Preprint arXiv:1603.09382 Wenlong Huang, Igor Mordatch, and Deepak Pathak. One policy to control them all: Shared modular policies for agent-agnostic control. In International Conference on Machine Learning, pp. 4455–4464, 2020. International Conference on Machine Learning Wenlong Huang, Pieter Abbeel, Deepak Pathak, and Igor Mordatch. Language models as zero-shot planners: Extracting actionable knowledge for embodied agents. Preprint arXiv:2201.07207, 2022. Preprint arXiv:2201.07207 David Yu-Tung Hui, Maxime Chevalier-Boisvert, Dzmitry Bahdanau, and Yoshua Bengio. Babyai 1.1. Preprint arXiv:2007.12770, 2020. Preprint arXiv:2007.12770 Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac, Carl Doersch, Catalin Ionescu, David Ding, Skanda Koppula, Daniel Zoran, Andrew Brock, Evan Shelhamer, et al. Perceiver IO: A general architecture for structured inputs & outputs. Preprint arXiv:2107.14795, 2021. Preprint arXiv:2107.14795 Michael Janner, Qiyang Li, and Sergey Levine. Offline reinforcement learning as one big sequence modeling problem. Advances in Neural Information Processing Systems, 34, 2021. Advances in Neural Information Processing Systems Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig. Scaling up visual and vision-language representation learning with noisy text supervision. In International Conference on Machine Learning, pp. 4904–4916, 2021. International Conference on Machine Learning Melvin Johnson, Orhan Firat, and Roee Aharoni. Massively multilingual neural machine translation. In Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 3874–3884, 2019. Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ronneberger, Kathryn Tunyasuvunakool, Russ Bates, Augustin Žídek, Anna Potapenko, et al. Highly accurate protein structure prediction with AlphaFold. Nature, 596(7873):583–589, 2021. Nature Lukasz Kaiser, Aidan N Gomez, Noam Shazeer, Ashish Vaswani, Niki Parmar, Llion Jones, and Jakob Uszkoreit. One model to learn them all. Preprint arXiv:1706.05137, 2017. Preprint arXiv:1706.05137 Anssi Kanervisto, Joonas Pussinen, and Ville Hautamäki. Benchmarking end-to-end behavioural cloning on video games. In IEEE conference on games (CoG), pp. 558–565, 2020. IEEE conference on games (CoG) Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models. Preprint arXiv:2001.08361, 2020. Preprint arXiv:2001.08361 Steven Kapturowski, Georg Ostrovski, John Quan, Remi Munos, and Will Dabney. Recurrent experience replay in distributed reinforcement learning. In International Conference on Learning Representations, 2018. International Conference on Learning Representations Zachary Kenton, Tom Everitt, Laura Weidinger, Iason Gabriel, Vladimir Mikulik, and Geoffrey Irving. Alignment of language agents. Preprint arXiv:2103.14659, 2021. Preprint arXiv:2103.14659 Nitish Shirish Keskar, Bryan McCann, Lav R Varshney, Caiming Xiong, and Richard Socher. CTRL: A conditional transformer language model for controllable generation. Preprint arXiv:1909.05858, 2019. Preprint arXiv:1909.05858 Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. Preprint arXiv:1412.6980, 2014. Preprint arXiv:1412.6980 Taku Kudo and John Richardson. SentencePiece: A simple and language independent subword tokenizer and detokenizer for neural text processing. In Annual Meeting of the Association for Computational Linguistics, Annual Meeting of the Association for Computational Linguistics pp. 66–71, 2018. Vitaly Kurin, Maximilian Igl, Tim Rocktäschel, Wendelin Boehmer, and Shimon Whiteson. My body is a cage: the role of morphology in graph-based incompatible control. Preprint arXiv:2010.01856, 2020. Preprint arXiv:2010.01856 Alex X Lee, Coline Manon Devin, Yuxiang Zhou, Thomas Lampe, Konstantinos Bousmalis, Jost Tobias Springenberg, Arunkumar Byravan, Abbas Abdolmaleki, Nimrod Gileadi, David Khosid, et al. Beyond pick-and-place: Tackling robotic stacking of diverse shapes. In Conference on Robot Learning, 2021. Conference on Robot Learning Alex X Lee, Coline Manon Devin, Jost Tobias Springenberg, Yuxiang Zhou, Thomas Lampe, Abbas Abdol-maleki, and Konstantinos Bousmalis. How to spend your robot time: Bridging kickstarting and offline reinforcement learning for vision-based robotic manipulation. Preprint arXiv:2205.03353, 2022. Preprint arXiv:2205.03353 Shuang Li, Xavier Puig, Chris Paxton, Yilun Du, Clinton Wang, Linxi Fan, Tao Chen, De-An Huang, Ekin Akyürek, Anima Anandkumar, Jacob Andreas, Igor Mordatch, Antonio Torralba, and Yuke Zhu. Pre-trained language models for interactive decision-making. Preprint arXiv:2202.01771, 2022a. Preprint arXiv:2202.01771 Yujia Li, David Choi, Junyoung Chung, Nate Kushman, Julian Schrittwieser, Rémi Leblond, Tom Eccles, James Keeling, Felix Gimeno, Agustin Dal Lago, et al. Competition-level code generation with AlphaCode. Preprint arXiv:2203.07814, 2022b. Preprint arXiv:2203.07814 Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. Preprint arXiv:1711.05101, 2017. Preprint arXiv:1711.05101 Kenneth Marino, Mohammad Rastegari, Ali Farhadi, and Roozbeh Mottaghi. Ok-VQA: A visual question answering benchmark requiring external knowledge. In IEEE Computer Vision and Pattern Recognition,pp. 3195–3204, 2019. IEEE Computer Vision and Pattern Recognition Jacob Menick, Maja Trebacz, Vladimir Mikulik, John Aslanides, Francis Song, Martin Chadwick, Mia Glaese, Susannah Young, Lucy Campbell-Gillingham, Geoffrey Irving, et al. Teaching language models to support answers with verified quotes. Preprint arXiv:2203.11147, 2022. Preprint arXiv:2203.11147 Margaret Mitchell, Simone Wu, Andrew Zaldivar, Parker Barnes, Lucy Vasserman, Ben Hutchinson, Elena Spitzer, Inioluwa Deborah Raji, and Timnit Gebru. Model cards for model reporting. In Proceedings of the conference on fairness, accountability, and transparency, pp. 220–229, 2019. Proceedings of the conference on fairness, accountability, and transparency Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. Human-level control through deep reinforcement learning. Nature, 518(7540):529–533, 2015. Nature Vernon Mountcastle. An organizing principle for cerebral function: the unit module and the distributed system. The mindful brain, 1978. The mindful brain Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, et al. WebGPT: Browser-assisted question-answering with human feedback. Preprint arXiv:2112.09332, 2021. Preprint arXiv:2112.09332 Aaron van den Oord, Sander Dieleman, Heiga Zen, Karen Simonyan, Oriol Vinyals, Alex Graves, Nal Kalchbrenner, Andrew Senior, and Koray Kavukcuoglu. WaveNet: A generative model for raw audio. Preprint arXiv:1609.03499, 2016. Preprint arXiv:1609.03499 Pedro A Ortega, Markus Kunesch, Grégoire Delétang, Tim Genewein, Jordi Grau-Moya, Joel Veness, Jonas Buchli, Jonas Degrave, Bilal Piot, Julien Perolat, et al. Shaking the foundations: delusions in sequence models for interaction and control. Preprint arXiv:2110.10819, 2021. Preprint arXiv:2110.10819 Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Preprint arXiv:2203.02155, 2022. Preprint arXiv:2203.02155 Simone Parisi, Aravind Rajeswaran, Senthil Purushwalkam, and Abhinav Gupta. The unsurprising effec-tiveness of pre-trained vision models for control. Preprint arXiv:2203.03580, 2022. Preprint arXiv:2203.03580 Vineel Pratap, Anuroop Sriram, Paden Tomasello, Awni Hannun, Vitaliy Liptchinsky, Gabriel Synnaeve, and Ronan Collobert. Massively multilingual ASR: 50 languages, 1 model, 1 billion parameters. Preprint arXiv:2007.03001, 2020. Preprint arXiv:2007.03001 Sébastien Racanière, Théophane Weber, David Reichert, Lars Buesing, Arthur Guez, Danilo Jimenez Rezende, Adrià Puigdomènech Badia, Oriol Vinyals, Nicolas Heess, Yujia Li, et al. Imagination-augmented agents for deep reinforcement learning. Advances in Neural Information Processing Systems, 30, 2017. Advances in Neural Information Processing Systems Jack W Rae, Sebastian Borgeaud, Trevor Cai, Katie Millican, Jordan Hoffmann, Francis Song, John Aslanides, Sarah Henderson, Roman Ring, Susannah Young, et al. Scaling language models: Methods, analysis & insights from training gopher. Preprint arXiv:2112.11446, 2021. Preprint arXiv:2112.11446 Scott Reed and Nando De Freitas. Neural programmer-interpreters. In International Conference on Learning Representations, 2016. International Conference on Learning Representations Machel Reid, Yutaro Yamada, and Shixiang Shane Gu. Can Wikipedia help offline reinforcement learning? Preprint arXiv:2201.12122, 2022. Preprint arXiv:2201.12122 Stuart Russell. Human compatible: Artificial intelligence and the problem of control. Penguin, 2019. Andrei A Rusu, Neil C Rabinowitz, Guillaume Desjardins, Hubert Soyer, James Kirkpatrick, Koray Human compatible: Artificial intelligence and the problem of control Kavukcuoglu, Razvan Pascanu, and Raia Hadsell. Progressive neural networks. Preprint arXiv:1606.04671, 2016. Preprint arXiv:1606.04671 Victor Sanh, Albert Webson, Colin Raffel, Stephen Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Arun Raja, Manan Dey, M Saiful Bari, Canwen Xu, Urmish Thakker, Shanya Sharma Sharma, Eliza Szczechla, Taewoon Kim, Gunjan Chhablani, Nihal Nayak, Debajyoti Datta, Jonathan Chang, Mike Tian-Jian Jiang, Han Wang, Matteo Manica, Sheng Shen, Zheng Xin Yong, Harshit Pandey, Rachel Bawden, Thomas Wang, Trishala Neeraj, Jos Rozen, Abheesht Sharma, Andrea Santilli, Thibault Fevry, Jason Alan Fries, Ryan Teehan, Teven Le Scao, Stella Biderman, Leo Gao, Thomas Wolf, and Alexander M Rush. Multitask prompted training enables zero-shot task generalization. In International Conference on Learning Representations, 2022. International Conference on Learning Representations Jürgen Schmidhuber. One big net for everything. Preprint arXiv:1802.08864, 2018. Preprint arXiv:1802.08864 Julian Schrittwieser, Ioannis Antonoglou, Thomas Hubert, Karen Simonyan, Laurent Sifre, Simon Schmitt, Arthur Guez, Edward Lockhart, Demis Hassabis, Thore Graepel, et al. Mastering atari, go, chess and shogi by planning with a learned model. Nature, 588(7839):604–609, 2020. Nature Piyush Sharma, Nan Ding, Sebastian Goodman, and Radu Soricut. Conceptual captions: A cleaned, hyper-nymed, image alt-text dataset for automatic image captioning. In Annual Meeting of the Association for Computational Linguistics, pp. 2556–2565, 2018. Annual Meeting of the Association for Computational Linguistics Noam Shazeer. Glu variants improve transformer. Preprint arXiv::2002.05202, 2020. Preprint arXiv::2002.05202 H Francis Song, Abbas Abdolmaleki, Jost Tobias Springenberg, Aidan Clark, Hubert Soyer, Jack W Rae, Seb Noury, Arun Ahuja, Siqi Liu, Dhruva Tirumala, et al. V-mpo: On-policy maximum a posteriori policy optimization for discrete and continuous control. In ICLR, 2020. ICLR Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(56): 1929–1958, 2014. Journal of Machine Learning Research Richard Sutton. The bitter lesson. Incomplete Ideas (blog), 13:12, 2019. Incomplete Ideas (blog) Yuval Tassa, Yotam Doron, Alistair Muldal, Tom Erez, Yazhe Li, Diego de Las Casas, David Budden, Abbas Abdolmaleki, Josh Merel, Andrew Lefrancq, et al. DeepMind control suite. Preprint arXiv:1801.00690, 2018. Preprint arXiv:1801.00690 Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. LaMDA: Language models for dialog applications. Preprint arXiv:2201.08239, 2022. Preprint arXiv:2201.08239 Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control. In International Conference on Intelligent Robots and Systems, pp. 5026–5033, 2012. International Conference on Intelligent Robots and Systems Maria Tsimpoukelli, Jacob L Menick, Serkan Cabi, SM Eslami, Oriol Vinyals, and Felix Hill. Multimodal few-shot learning with frozen language models. Advances in Neural Information Processing Systems, pp. 200–212, 2021. Advances in Neural Information Processing Systems Saran Tunyasuvunakool, Alistair Muldal, Yotam Doron, Siqi Liu, Steven Bohez, Josh Merel, Tom Erez, Timothy Lillicrap, Nicolas Heess, and Yuval Tassa. dm_control: Software and tasks for continuous control. Software Impacts, 6:100022, 2020. Software Impacts Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in Neural Information Processing Systems, 30, 2017. Advances in Neural Information Processing Systems Zirui Wang, Jiahui Yu, Adams Wei Yu, Zihang Dai, Yulia Tsvetkov, and Yuan Cao. Simvlm: Simple visual language model pretraining with weak supervision. Preprint arXiv:2108.10904, 2021. Preprint arXiv:2108.10904 Ziyu Wang, Alexander Novikov, Konrad Zolna, Josh S Merel, Jost Tobias Springenberg, Scott E Reed, Bobak Shahriari, Noah Siegel, Caglar Gulcehre, Nicolas Heess, et al. Critic regularized regression. Advances in Neural Information Processing Systems, 33:7768–7778, 2020. Advances in Neural Information Processing Systems Jason Wei, Maarten Bosma, Vincent Y Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M Dai, and Quoc V Le. Finetuned language models are zero-shot learners. Preprint arXiv:2109.01652, 2021. Preprint arXiv:2109.01652 Laura Weidinger, John Mellor, Maribeth Rauh, Conor Griffin, Jonathan Uesato, Po-Sen Huang, Myra Cheng, Mia Glaese, Borja Balle, Atoosa Kasirzadeh, et al. Ethical and social risks of harm from language models. Preprint arXiv:2112.04359, 2021. Preprint arXiv:2112.04359 Yuxin Wu and Kaiming He. Group normalization. In European Conference on Computer Vision, pp. 3–19, 2018. European Conference on Computer Vision Tianhe Yu, Deirdre Quillen, Zhanpeng He, Ryan Julian, Karol Hausman, Chelsea Finn, and Sergey Levine. Meta-World: A benchmark and evaluation for multi-task and meta reinforcement learning. In Conference on Robot Learning, pp. 1094–1100, 2020. Conference on Robot Learning Qinqing Zheng, Amy Zhang, and Aditya Grover. Online decision transformer. Preprint arXiv:2202.05607, 2022. Preprint arXiv:2202.05607 Konrad Zolna, Alexander Novikov, Ksenia Konyushkova, Caglar Gulcehre, Ziyu Wang, Yusuf Aytar, Misha Denil, Nando de Freitas, and Scott Reed. Offline learning from demonstrations and unlabeled experience. Preprint arXiv:2011.13885, 2020. Preprint arXiv:2011.13885 Konrad Zolna, Scott Reed, Alexander Novikov, Sergio Gómez Colmenarejo, David Budden, Serkan Cabi, Misha Denil, Nando de Freitas, and Ziyu Wang. Task-relevant adversarial imitation learning. In Conference on Robot Learning, pp. 247–263, 2021. Conference on Robot Learning Supplementary Material A Model card We present a model card for Gato in Table 4. 4. Table 4: Gato Model Card. We follow the framework proposed in (Mitchell et al., 2019). Gato Model Card. (Mitchell et al., 2019). B Agent Data Tokenization Details In this section we provide additional details on our tokenization schemes. Our agent data is sequenced as follows: • Episodes are presented to the agent in order of time (timesteps). Episodes • Timesteps in turn are presented in the following order: Timesteps – Observations ([y1:k, x1:m, z1:n]) are ordered lexicographically by key, each item is sequenced as follows: – Observations y k, x m, z n ∗ Text tokens (y1:k) are in the same order as the raw input text. y k ∗ Image patch tokens (x1:m) are in raster order. x m ∗ Tensors (z1:n) (such as discrete and continuous observations) are in row-major order. z n – Separator ('|'); a designated separator token is provided after observations. Separator | – Actions (a1:A) are tokenized as discrete or continuous values and in row-major order. Actions a A A full sequence of tokens is thus given as the concatenation of data from T timesteps: where L = T(k + m + n + 1 + A) is the total number of tokens. Each floating point element of tensors in the observation sequence is mu-law companded as in WaveNet (Oord et al., 2016): (Oord et al., 2016): with parameters µ = 100 and M = 256. (If the floating-point tensor is in the action set, we do not need to compand the elements in the sequence because actions are only defined in the range \[ 1, 1\] for all our environments.) All the elements are subsequently clipped so that they fall in the set \[ 1, 1\]. Finally, they are discretized using bins of uniform width on the domain \[ 1,1\]. We use 1024 bins and shift the resulting integers so they are not overlapping with the ones used for text tokens. The tokenized result is therefore a sequence of integers within the range of \[32000, 33024). See Figure 14 and Figure 15 for visualizations of tokenizing and sequencing values (both discrete and con-tinuous) and images. See Section C for details about local position encodings referenced in the figures. 14 15 C C Model Architecture C.1 Transformer Hyperparameters The transformer hyperparameters of Gato are presented in Table 5. We also list the hyperparameters of smaller architecture variants used in Section 5. 5. 5. C.2 Embedding Function The ResNet block uses the v2 architecture (He et al., 2016b), contains GroupNorm (Wu & He, 2018) with 32 groups instead of LayerNorm (Ba et al., 2016), and GELU (Hendrycks & Gimpel, 2016) activation functions instead of RELU. The block is diagrammed in Figure 16. (He et al., 2016b), (Wu & He, 2018) (Ba et al., 2016), (Hendrycks & Gimpel, 2016) 16. C.3 Position Encodings After tokens are mapped into token embeddings, two position encodings are added to the token embeddings (when applicable) to provide temporal and spatial information to the model. These are described below. Patch Position Encodings These position encodings convey information about a patch’s global position within the image from which the patch was extracted. First, the relative row and column intervals of the patch are calculated by normalizing the patch’s pixel intervals by the image resolution. The row and column normalized intervals are then quantized into a vocabulary size (we use 128) and are used to index a row and column table of learnable position encodings. The method in which the quantized row and column intervals are converted into indices depends on whether we are training or evaluating the model: during training a random index is uniformly sampled from the quantized interval, while during evaluation we deterministically take the (rounded) mean of the interval. Once row and column position encoding are retrieved from the embedding table, they are added onto the token embedding produced by the resnet embedding function, as described previously. To more concretely demonstrate this process, we provide an example in Figure [17.](#_bookmark144) We will follow the process with the patch highlighted in red on the left of the subfigure. The image is of resolution 80 64 and each patch is 16 16, meaning there are 5 4 = 20 patches total. The highlighted patch starts at pixel row interval \[16*,* 32\] and pixel column interval \[32*,* 64\]. Normalized, the row interval is therefore \[0*.*25*,* 0*.*5\] and the column interval is \[0*.*4*,* 0*.*6\]. We then separately quantize the intervals into 128 uniformly spaced bins, with the resulting quantized row interval being \[32*,* 64\] and the quantized column interval being \[51*,* 77\]. During training, we uniformly sample integers between the quantized row intervals, whereas during testing we would use the means, which are index 48 for row position and index 64 for column position. The row and column positions are finally used to index separate row and column position encoding tables to produce learnable embeddings which are added onto the corresponding patch token embedding. Local Observation Position Encodings The local observation position encoding adds positional information about where observation tokens are positioned within the local time-step they were an element of. First, we reiterate that, during tokenization, for each time-step all elements of the observation set are tokenized into sequences and concatenated into an observation sequence. Each token in this observation sequence is given an index which corresponds to the sequence order, i.e. the first token is 0 and the last is the length of the observation sequence minus one. After embedding, for any tokens that were a part of an observation set, the corresponding observation token index is used to index an embedding table of learnable position encodings, with one embedding for every possible observation token index (in practice we simply set the table size to a large value like 512). / The position encoding is then added onto the observation token embedding to produce the final token embedding. Note that all action tokens are given the same position encoding regardless of their position in the time-step sequence. We illustrate an example of this process in Figure 18. 18. D Pretraining Setup Optimizer: For all models we use the AdamW (Loshchilov & Hutter, 2017) optimizer with a linear warm-up and cosine schedule decay. The linear warmup lasts for 15*,* 000 steps, starting from a learning rate of 1e-7 and ending at a different maximum learning rate depending on the model (see Table 6). This learning rate is then cosine decayed by a factor 10x over 1,000,000 steps. The AdamW optimizer has parameters β1 = 0*.9, β2 = 0.*95 and ϵ = 1e-8. We use a batch size of 512 and a sequence length of 1024 tokens for all models. Optimizer: (Loshchilov & Hutter, 2017) 6). β 9, β ϵ Regularization: We train with an AdamW weight decay parameter of 0.1. Additionally, we use stochastic depth (Huang et al., 2016) during pretraining, where each of the transformer sub-layers (i.e. each Multi-Head Attention and Dense Feedforward layer) is skipped with a probability of 0.1. Regularization: (Huang et al., 2016) E Fine-tuning Setup Optimizer: For all models we use the Adam (Kingma & Ba, 2014) optimizer with a constant learning rate of 1e-5. The Adam optimizer has parameters β1 = 0*.9, β2 = 0.*95 and ϵ = 1e-8. We use a batch size of 64 and a sequence length of 1024 tokens for all models. We train for 10,000 gradient steps. Optimizer: (Kingma & Ba, 2014) β 9, β ϵ Regularization: We use dropout (Srivastava et al., 2014) with a rate of 0.1. Regularization: (Srivastava et al., 2014) Evaluation: We evaluate agent every 100 learning steps. Each evaluation reports the average of 10 runs of a given checkpoint. The moving average of 5 such scores is computed (to gather 50 runs together). The final fine-tuning performance is defined as the maximum of these smoothed scores. Evaluation: Datasets: We generated data for the fine-tuning tasks the same way we did for the other tasks (see Section 3.1 for details). Instead of using all the data for a fine-tuning task, we discarded all but 2000 best episodes (leading to the highest returns). The fine-tuning datasets were created in the following way. We randomly took 1000 episodes (out of 2000 preselected episodes), then a subset of 100 episodes from the selected episodes, then 10, 5, 3, and finally a single episode. We repeated this procedure 3 times to obtain 3 series of cascading subsets for each task. Each subset is used to conduct one fine-tuning experiment, and each is reported on our plots in Section 5.2 as a separate point. Datasets: 5.2 Task settings: We have not altered any of the tasks and used their canonical versions. As 3 out of 4 tasks are open sourced, they do not need further explanation. For the fourth task, DMLab order_of_apples_forage_simple, the goal is to collect apples in the right order, green ones first followed by the gold one. Task settings: F Data Collection Details F.1 Atari We collect two separate sets of Atari environments. The first (that we refer to as ALE Atari) consists of 51 canonical games from the Arcade Learning Environment (Bellemare et al., 2013). The second (that we refer to as ALE Atari Extended) is a set of alternative games3 with their game mode and difficulty randomly set at the beginning of each episode. (Bellemare et al., 2013). 3 For each environment in these sets we collect data by training a Muesli (Hessel et al., 2021) agent for 200M total environment steps. We record approximately 20,000 random episodes generated by the agent during training. (Hessel et al., 2021) F.2 Sokoban Sokoban is a planning problem (Racanière et al., 2017), in which the agent has to push boxes to target locations. Some of the moves are irreversible and consequently mistakes can render the puzzle unsolvable. Planning ahead of time is therefore necessary to succeed at this puzzle. We use a Muesli (Hessel et al., 2021) agent to collect training data. (Racanière et al., 2017), (Hessel et al., 2021) F.3 BabyAI BabyAI is a gridworld environment whose levels consist of instruction-following tasks that are described by a synthetic language. We generate data for these levels with the built-in BabyAI bot. The bot has access to extra information which is used to execute optimal solutions, see Section C in the appendix of (Chevalier-Boisvert et al., 2018) for more details about the bot. We collect 100,000 episodes for each level. (Chevalier-Boisvert et al., 2018) F.4 DeepMind Control Suite The DeepMind Control Suite (Tunyasuvunakool et al., 2020; Tassa et al., 2018) is a set of physics-based simulation environments. For each task in the control suite we collect two disjoint sets of data, one using only state features and another using only pixels. We use a D4PG (Barth-Maron et al., 2018) agent to collect data from tasks with state features, and an MPO (Abdolmaleki et al., 2018) based agent to collect data using pixels. unyasuvunakool et al 2020; Tassa et al., 2018) (Barth-Maron et al., 2018) (Abdolmaleki et al., 2018) We also collect data for randomized versions of the control suite tasks with a D4PG agent. These versions randomize the actuator gear, joint range, stiffness, and damping, and geom size and density. There are two difficulty settings for the randomized versions. The small setting scales values by a random number sampled from the union of intervals [0*.9,* 0*.95] ∪ [1.05,* 1*.1]. The large setting scales values by a random number sampled from the union of intervals [0.6,* 0*.8] ∪ [1.2,* 1*.*4]. 9 95] ∪ 05 1]. The large setting scales values by a random number sampled from the union of intervals [0 6 8] ∪ 2 F.5 DeepMind Lab DeepMind Lab (Beattie et al., 2016) is a first-person 3D environment designed to teach agents 3D vision from raw pixel inputs with an egocentric viewpoint, navigation, and planning. (Beattie et al. 2016) We trained an IMPALA (Espeholt et al., 2018) agent jointly on a set of 18 parent DM Lab levels that generate maps procedurally for each new episode. Data was collected by executing the agent on these 18 levels, as well as an additional set of 237 levels handcrafted to test a diverse set of skills. (Espeholt et al., 2018) The 18 parent levels are characterized by high diversity of generated maps. The difference between the levels is rooted in hyper-parameters used in a generation process. These hyper-parameters control high-level characteristics such as types of structures spawned, difficulty of language instructions, or presence of specific tools. The parent levels were developed to improve performance of RL agents trained online on them. In contrast to the parent levels, each of the additional handcrafted 237 levels uses almost the same map, and the main differences between instances of the same level map are aesthetics such as colors of walls or lighting conditions. The maps are not procedurally generated and were designed to test a diverse set of skills such as walking up stairs or using specific tools. They are similar to levels presented in Figure 3, Figure 7 and Figure 8 in aforementioned paper by Beattie et al. (2016). not Beattie et al. (2016). Additional information on the 18 parent levels (and their relation to the other levels) is presnted in details in the NeurIPS Workshop talk A Methodology for RL Environment Research by Daniel Tanis4. A Methodology for RL Environment Research 4 In total, we collected data for 255 levels from the DeepMind Lab (18 parent levels and 237 handcrafted levels), 254 of which were used while training Gato. The remaining level was used for out of distribution evaluation. F.6 Procgen Benchmark Procgen (Cobbe et al., 2020) is a suite of 16 procedurally generated Atari-like environments, which was proposed to benchmark sample efficiency and generalization in reinforcement learning. Data collection was done while training a R2D2 (Kapturowski et al., 2018) agent on each of the environments. We used the hard difficulty setting for all environments except for maze and heist, which we set to easy. (Cobbe et al., 2020) (Kapturowski et al., 2018) F.7 Modular RL Modular RL (Huang et al., 2020) is a collection of MuJoCo (Todorov et al., 2012) based continuous control environments, composed of three sets of variants of the OpenAI Gym (Brockman et al., 2016) Walker2d-v2, Humanoid-v2, and Hopper-v2. Each variant is a morphological modification of the original body: the set of morphologies is generated by enumerating all possible subsets of limbs, and keeping only those sets that a) contain the torso, and b) still form a connected graph. This results in a set of variants with different input and output sizes, as well as different dynamics than the original morphologies. We collected data by training a single morphology-specific D4PG agent on each variant for a total of 140M actor steps, this was done for 30 random seeds per variant. (Huang et al., 2020) odorov et al., 2012) (Brockman et al., 2016) F.8 DeepMind Manipulation Playground The DeepMind Manipulation Playground (Zolna et al., 2021) is a suite of MuJoCo based simulated robot tasks. We collect data for 4 of the Jaco tasks (box, stack banana, insertion, and slide) using a Critic-Regularized Regression (CRR) agent (Wang et al., 2020) trained from images on human demonstrations. The collected data includes the MuJoCo physics state, which is we use for training and evaluating Gato. (Zolna et al., 2021) (Wang et al., 2020) F.9 Meta-World Meta-World (Yu et al., 2020) is a suite of environments5 for benchmarking meta-reinforcement learning and multi-task learning. We collect data from all train and test tasks in the MT50 mode by training a MPO agent (Abdolmaleki et al., 2018) with unlimited environment seeds and with access to state of the MuJoCo physics engine. The collected data also contains the MuJoCo physics engine state. u et al., 2020) 5 (Abdolmaleki et al., 2018) G Real robotics evaluation details In the real world, control is asynchronous; physics does not wait for computations to finish. Thus, inference latency is a concern for evaluating a large model for real world tasks. In robotics, a fast control rate is thought to be critical for reacting to dynamic phenomena. The robot setup for RGB stacking has a 20Hz control rate (0.05 second timestep) by design. In order to reach an acceptable margin of latency, we modified inference at evaluation time by shortening the context length to 1. We also implemented a parallel sampling scheme where all the action tokens are zeroed out in the input sequences during training so we can sample all tokens corresponding to a robot action in a single model inference step instead of autoregressively as it’s done in other domains. We found that the 1.18B parameter model was able to run on the hardware accelerators in our robots (NVidia GeForce RTX 3090s), but still overran the 20Hz control rate by a small amount (~0.01 seconds). We use the sparse reward function described in Lee et al. (2021) for data filtering. We only select trajectories with final task success; that is, a sparse reward of 1 on the final timestep. Lee et al. (2021) final H Skill Mastery architecture The numbers reported for the Skill Mastery benchmark were collected by executing a model zero-shot that used an earlier version of the Gato architecture. Instead of the ResNet patch embedding, a similar architecture using a local transformer was used to embed image patch tokens. The local position embeddings and patch position embeddings were not used. These changes were implemented and found to improve Gato’s performance after the pretraining data was changed (as we decided to focus on Skill Generalization instead of Skill Mastery challenge), which is why they are presented as the final architecture of our full model. I Additional robotics ablations We conducted a series of ablations in simulation to better understand the effect of diverse pretraining data in the robotics domain (see Figure 19). We included the same baselines as in Section 5.2, selecting the 364M parameter size variant, as well as an additional baseline trained with control suite data only. The DM Control-only agent is superior to the base Gato at zero-shot transfer and with a lot of fine-tuning data, suggesting that Gato may not be using the representations learned from the text-based datasets when adapting to robotics tasks. The same domain only agent performs the best overall, matching the CRR baseline at 1 fine-tuning episode and outperforming it with more data, suggesting that Gato at current scale can trade its generalization capacity for data-efficient and effective few-shot adaptation. 19). 5.2, J Attention visualization To render the transformer attention weights, we retrieved the cross-attention logits, a tensor with dimension (H, T, T ) where H is the number of heads and T is the number of tokens in a sequence. The (h, i, j)th entry of this matrix can be interpreted as the amount that head h attends to token j from token i. Due to Gato’s image tokenization scheme, there are multiple tokens per timestep. Therefore to render the attention for a particular timestep, we took the sub-matrix that corresponds to that timestep. We then applied a softmax over the rows of this matrix to normalize the relevant values. Because we are only interested in attention to the previous tokens, we excluded the diagonal by setting it to negative infinity before softmax. H, T, T H T h, i, j h j i To measure the importance of each patch, we averaged the attention weights over the corresponding column. Because Gato uses a causal transformer, the attention matrix is lower triangular, so the mean was only considered over the sub-column below the diagonal of the matrix. This corresponds to the average attention paid to particular patch over a whole timestep. Using this method, we found the attention maps at the first layer the transformer to be most interpretable, agreeing with the findings of Abnar & Zuidema (2020). Certain heads clearly track task-specific entities and regions of the image. Figure 20 shows the attention maps for manually selected heads at the first layer for several tasks. Abnar & Zuidema (2020). 20 K Detailed results for specialist Meta-World agent The specialist Meta-World agent described in Section 5.5 achieves 96.6% success rate averaged over all 50 Meta-World tasks. The detailed success rates are presented in Table 7. We evaluated agent 500 times for each task. 5.5 7. L Per-domain results for Gato We describe performance of Gato for simulated control tasks in Section 4.1. In Table 8, we present normalized per-domain results. We evaluated agent 50 times for each task. 4.1. 8, This paper is available on arxiv under CC by 4.0 Deed (Attribution 4.0 International) license. This paper is available on arxiv under CC by 4.0 Deed (Attribution 4.0 International) license. available on arxiv
