The ability to act in multiple environments and transfer previous knowledge to new situations can be considered a critical aspect of any intelligent agent. Towards this goal, we define a novel method of multitask and transfer learning that enables an autonomous agent to learn how to behave in multiple tasks simultaneously, and then generalize its knowledge to new domains. This method, termed "Actor-Mimic", exploits the use of deep reinforcement learning and model compression techniques to train a single policy network that learns how to act in a set of distinct tasks by using the guidance of several expert teachers. We then show that the representations learnt by the deep policy network are capable of generalizing to new tasks with no prior expert guidance, speeding up learning in novel environments. Although our method can in general be applied to a wide range of problems, we use Atari games as a testing environment to demonstrate these methods.

From the early days of computing, games have been important testbeds for studying how well machines can do sophisticated decision making. In recent years, machine learning has made dramatic advances with artificial agents reaching superhuman performance in challenge domains like Go, Atari, and some variants of poker. As with their predecessors of chess, checkers, and backgammon, these game domains have driven research by providing sophisticated yet well-defined challenges for artificial intelligence practitioners. We continue this tradition by proposing the game of Hanabi as a new challenge domain with novel problems that arise from its combination of purely cooperative gameplay with two to five players and imperfect information. In particular, we argue that Hanabi elevates reasoning about the beliefs and intentions of other agents to the foreground. We believe developing novel techniques for such theory of mind reasoning will not only be crucial for success in Hanabi, but also in broader collaborative efforts, especially those with human partners. To facilitate future research, we introduce the open-source Hanabi Learning Environment, propose an experimental framework for the research community to evaluate algorithmic advances, and assess the performance of current state-of-the-art techniques. 6 One such equilibrium occurs when players do not intentionally communicate information to other players, and ignore what other players tell them (historically called a pooling equilibrium in pure signalling games [15], or a babbling equilibrium in later work using cheap talk [16]). In this case, there is no incentive for a player to start communicating because they will be ignored, and there is no incentive to pay attention to other players because they are not communicating.7 In pure signalling games where actions are purely communicative, policies are often referred to as communication protocols. Though Hanabi is not such a pure signalling game, when we want to emphasize the communication properties of an agent's policy we will still refer to its communication protocol. 8 We use the word convention to refer to the parts of a communication protocol or policy that interrelate. Technically, these can be thought of as constraints on the policy to enact the convention.

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.

The ability for an agent to localize itself within an environment is crucial for many real-world applications. For unknown environments, Simultaneous Localization and Mapping (SLAM) enables incremental and concurrent building of and localizing within a map. We present a new, differentiable architecture, Neural Graph Optimizer, progressing towards a complete neural network solution for SLAM by designing a system composed of a local pose estimation model, a novel pose selection module, and a novel graph optimization process. The entire architecture is trained in an end-to-end fashion, enabling the network to automatically learn domain-specific features relevant to the visual odometry and avoid the involved process of feature engineering. We demonstrate the effectiveness of our system on a simulated 2D maze and the 3D ViZ-Doom environment.

We investigate the use of prior knowledge of human and animal movement to learn reusable locomotion skills for real legged robots. Our approach builds upon previous work on imitating human or dog Motion Capture (MoCap) data to learn a movement skill module. Once learned, this skill module can be reused for complex downstream tasks. Importantly, due to the prior imposed by the MoCap data, our approach does not require extensive reward engineering to produce sensible and natural looking behavior at the time of reuse. This makes it easy to create well-regularized, task-oriented controllers that are suitable for deployment on real robots. We demonstrate how our skill module can be used for imitation, and train controllable walking and ball dribbling policies for both the ANYmal quadruped and OP3 humanoid. These policies are then deployed on hardware via zero-shot simulation-to-reality transfer. Accompanying videos are available at https://bit.ly/robot-npmp.

Many real-world applications such as robotics provide hard constraints on power and compute that limit the viable model complexity of Reinforcement Learning (RL) agents. Similarly, in many distributed RL settings, acting is done on unaccelerated hardware such as CPUs, which likewise restricts model size to prevent intractable experiment run times. These "actor-latency" constrained settings present a major obstruction to the scaling up of model complexity that has recently been extremely successful in supervised learning. To be able to utilize large model capacity while still operating within the limits imposed by the system during acting, we develop an "Actor-Learner Distillation" (ALD) procedure that leverages a continual form of distillation that transfers learning progress from a large capacity learner model to a small capacity actor model. As a case study, we develop this procedure in the context of partially-observable environments, where transformer models have had large improvements over LSTMs recently, at the cost of significantly higher computational complexity. With transformer models as the learner and LSTMs as the actor, we demonstrate in several challenging memory environments that using Actor-Learner Distillation recovers the clear sample-efficiency gains of the transformer learner model while maintaining the fast inference and reduced total training time of the LSTM actor model.

No abstract

The recent phenomenal success of language models has reinvigorated machine learning research, and large sequence models such as transformers are being applied to a variety of domains. One important problem class that has remained relatively elusive however is purposeful adaptive behavior. Currently there is a common perception that sequence models "lack the understanding of the cause and effect of their actions" leading them to draw incorrect inferences due to auto-suggestive delusions. In this report we explain where this mismatch originates, and show that it can be resolved by treating actions as causal interventions. Finally, we show that in supervised learning, one can teach a system to condition or intervene on data by training with factual and counterfactual error signals respectively.