Reproduce “Smooth Action Chunking Via Movement-Primitive Representations for Imitation Learning” 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
Shota Nomura, Takayuki Osa, Yasuhiro Kato, Kazuki Ota, Naoki Shitanda, Tatsuya Harada
The authors introduce a method that predicts smooth trajectory-basis coefficients instead of raw action chunks, then adjusts them at inference to connect each new chunk to the previous command. With ACT and DiT, it reduces within-chunk jerk; on Fold Cloth with ACT, jerk falls from 434.2 ± 19.1 to 33.4 ± 1.4. Hardware tests found no significant success change; power was limited.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.