Abstract
Deep reinforcement learning has shown lots of success in closed, well-defined domains such as games (Chess, Go, StarCraft). The next frontier is real-world scenarios, where setups are numerous and varied. For this, agents need to learn the underlying environment dynamics, so as to robustly generalise to conditions that differ from those they were trained on. Model-based reinforcement learning algorithms, such as MuZero or Dreamer, aim to accomplish this by learning a world model. However, leveraging a world model has not yet consistently shown greater generalisation capabilities compared to model-free alternatives. In this work, we propose improving the data efficiency and generalisation capabilities of MuZero by explicitly incorporating the symmetries of the environment in its world-model architecture. We prove that, so long as the neural networks used by MuZero are equivariant to a particular symmetry group acting on the environment, the entirety of MuZero's action-selection algorithm will also be equivariant to that group. As such, Equivariant MuZero is guaranteed to behave symmetrically in symmetrically-transformed states, and will hence be more data-efficient when learning its world models. We evaluate Equivariant MuZero on procedurally-generated MiniPacman and on Chaser from the ProcGen suite: training on a set of mazes, and then testing on unseen rotated versions, demonstrating the benefits of equivariance. We verify that our improvements hold even when only some of the components of Equivariant MuZero obey strict equivariance, which highlights the robustness of our construction.
Authors
Andreea Deac*, Theophane Weber, George Papamakarios
- *
- External author
Venue
Transactions on Machine Learning Research (TMLR)