Reproduce “Evaluating model-based planning and planner amortization for continuous control” 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
Arunkumar Byravan, Leonard Hasenclever, Piotr Trochim, Mehdi Mirza, Alessandro Davide Ialongo, Yuval Tassa, Jost Tobias Springenberg, Abbas Abdolmaleki and 3 more
The authors test whether model-based control improves learning across varied locomotion goals and whether planning can later be removed. They guide short-horizon planning with a learned policy, then train that policy using behavioral cloning and MPO. In simulated go-to-target-pose tasks, the combined method improves learning, especially for OP3; results measure environment interactions, not total computation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.