Reproduce “Continuous-Discrete Reinforcement Learning for Hybrid Control in Robotics” 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
Michael Neunert, Abbas Abdolmaleki, Markus Wulfmeier, Thomas Lampe, Jost Tobias Springenberg, Roland Hafner, Francesco Romano, Jonas Buchli and 2 more
Hybrid MPO lets a policy choose continuous actions and discrete commands together. It models continuous actions with Gaussians and discrete commands with categoricals, limiting changes to each. On Sawyer, it learned to grasp and lift a cube where standard MPO plateaued after reaching. In peg-in-hole, its average final reward was about 2,750, above two fixed-mode choices in that setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.