Reproduce “Collective Learning for Unified Robotic Manipulation via Predictive Dynamics Adaptation” 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
Shaocheng Liu, Yuan Meng, Zhenshan Bing, Alois Knoll
PACE separates task intent from robot-dependent execution behavior in collective robot learning. It uses trajectory prediction to shape policy representations without online planning. In Meta-World simulations, it achieves 93.21% ± 2.81% success on four unseen robots, versus 81.75% ± 5.99% for Transformer behavior cloning. On a physical KUKA robot, it averages 79.2% across six tasks, with weaker contact-rich performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.