Reproduce “Spectrally Routed Diffusion Policy for Multi-Task Robotic Manipulation” 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
Mingcong Li, Yihan Zhou, Jiayu Qu, Xueyang Guo, Yilin Mo
SR-DP combines spatially informative visual tokens with a shared diffusion policy whose selected denoiser layers receive task-conditioned adjustments through a frozen SVD basis. On six jointly trained RoboTwin tasks, it achieves 68.4% mean success, versus 46.6% for multi-task VO-DP and above the specialist average, but does not outperform specialists on every task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.