Reproduce “DemoStream: Batched Task-and-Motion-Plan Optimization for Automated Learning from Demonstration” 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
Caelan Garrett, Neel Jawale, Fabio Ramos
DemoStream generates robot demonstrations by searching feasible task-and-motion plans for low-cost candidates. Its batch samplers evaluate alternatives together, and its optimizer selects bindings using task-space or joint-space costs. In Stack-Three simulation, task-space cost with T-PRM yielded 76% BC-RNN policy success, versus 34% with joint-space cost and 0% with zero cost. Real-world results cover three tasks, with human resets and labeling.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.