Reproduce “TOPS: Reshaping Flow Source and Transport for Multi-Task Robot Manipulation” 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

Tuo An, Chuhao Zhou, Jiahua Dong, Jingliang Li, Jiaqi Bai, Jindou Jia, Jianfei Yang

TOPS changes how a shared flow-matching robot policy generates actions across tasks. It reverses demonstrated actions through a warm-up policy to fit task-specific sources, then schedules movement toward expert actions using covariance mismatch. On LIBERO-10, TOPS reached 69.67% success with two velocity-network evaluations, versus 53.44% for Standard Flow Matching, but reported latency excludes offline construction costs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.