Reproduce “ActionCodec2: A Streaming Action Tokenizer” 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
Zibin Dong, Yicheng Liu, Shiduo Zhang, Tianyuan Yuan, Baijun Ye, Jinqiu Wang, Hang Zhao
ActionCodec2 streams actions as tokens and chooses each motion step relative to the decoder’s reconstructed position, correcting residual error rather than quantizing increments independently. This bounds reconstructed trajectory error for feasible motions within specified tolerance, but does not guarantee physical execution. On four zero-shot DROID–Franka task suites, ACodec2+ achieved 73.9% task progress versus FAST’s 50.8%, measured as prescribed milestones attained.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.