Reproduce “Augmenting learning using symmetry in a biologically-inspired domain” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shruti Mishra, Abbas Abdolmaleki, Arthur Guez, Piotr Trochim, Doina Precup
The authors augment an actor–critic learner with mirrored states and actions from a simulated quadruped’s experience, using symmetry to create training examples without new environment interactions. They apply the method to MPO on walk and run tasks in a data-limited regime. Augmented training performs better, typically in fewer episodes; this preliminary result is limited to simulation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.