Reproduce “PaceVLA: Adaptive Temporal Action Quantization in Vision-Language-Action Models” 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
Hojin Jeon, Jimin Lee, Taekwan Kim, Nayoung Oh, Taeyoung Kim, Myungkyu Koo, Jinmo Koo, Joonwoo Ahn and 1 more
PaceVLA lets a vision-language-action policy adjust pace across task phases, selecting both command granularity and prediction horizon. It uses a quantization judge and context-conditioned routing among action decoders. On ALLEX conveyor package handling, GR00T N1.5 with PaceVLA reports 6.3 successful packages per minute and 94% success, versus 3.6 and 86%; added inference latency and perception errors remain limitations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.