There is a widespread intuition that model-based control methods should be able to surpass the data efficiency of model-free approaches. In this paper we attempt to evaluate this intuition on various challenging locomotion tasks. We take a hybrid approach, combining model predictive control (MPC) with a learned model and model-free policy learning; the learned policy serves as a proposal for MPC. We find that well-tuned model-free agents are strong baselines even for high DoF control problems but MPC with learned proposals and models (trained on the fly or transferred from related tasks) can significantly improve performance and data efficiency in hard multi-task/multi-goal settings. Finally, we show that it is possible to distil a model-based planner into a policy that amortizes the planning computation without any loss of performance. Videos of agents performing different tasks can be seen at https://sites.google.com/view/mbrl-amortization/home.

In the pursuit of artificial general intelligence, our most significant measurement of progress is an agent's ability to achieve goals in a wide range of environments. Existing platforms for constructing such environments are typically constrained by the technologies they are founded on, and are therefore only able to provide a subset of scenarios necessary to evaluate progress. To overcome these shortcomings, we present our use of Unity, a widely recognized and comprehensive game engine, to create more diverse, complex, virtual simulations. We describe the concepts and components developed to simplify the authoring of these environments, intended for use predominantly in the field of reinforcement learning. We also introduce a practical approach to packaging and re-distributing environments in a way that attempts to improve the robustness and reproducibility of experiment results. To illustrate the versatility of our use of Unity compared to other solutions, we highlight environments already created using our approach from published papers. We hope that others can draw inspiration from how we adapted Unity to our needs, and anticipate increasingly varied and complex environments to emerge from our approach as familiarity grows.

A MuJoCo wrapper provides convenient bindings to functions and data structures. The PyMJCF and Composer libraries enable procedural model manipulation and task authoring. The Control Suite is a fixed set of tasks with standardised structure, intended to serve as performance benchmarks. The Locomotion framework provides high-level abstractions and examples of locomotion tasks. A set of configurable manipulation tasks with a robot arm and snap-together bricks is also included.

Invariances to translation, rotation and other spatial transformations are a hallmark of the laws of motion, and have widespread use in the natural sciences to reduce the dimensionality of systems of equations. In supervised learning, such as in image classification tasks, rotation, translation and scale invariances are used to augment training datasets. In this work, we use data augmentation in a similar way, exploiting symmetry in the quadruped domain of the DeepMind control suite (Tassa et al. 2018) to add to the trajectories experienced by the actor in the actor-critic algorithm of Abdolmaleki et al. (2018). In a data-limited regime, the agent using a set of experiences augmented through symmetry is able to learn faster. Our approach can be used to inject knowledge of invariances in the domain and task to augment learning in robots, and more generally, to speed up learning in realistic robotics applications.