Reproduce “R3D: Revisiting 3D Policy 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
Zhengdong Hong, Shenrui Wu, Haozhe Cui, Boyi Zhao, Ran Ji, Yiyang He, Hangxing Zhang, Zundong Ke and 3 more
R3D combines a transformer-based 3D point-cloud encoder that retains spatial features with a diffusion action decoder. The authors identify an inference-time exponential moving average bug that left batch-normalization statistics unsynchronized, and report that 3D augmentation helps address instability and overfitting. Across six real-world tasks on one xArm6, it averages 60.7% success, versus 44.0% for Diffusion Policy and 29.7% for DP3.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.