Reproduce “Towards Efficient Robotic Manipulation Models with Self-Recursive Pruning” 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
Zijia Chen, Yuenan Hou, Yu Li, Weijie Li, Li Liu
LCAM prunes robotic policies by scoring connections using relative weight contribution, demonstration activations, and sensitivity of action-prediction loss, then increasing sparsity gradually with recalibration. On LIBERO-Object, OpenVLA-7B achieved 84.0% rollout success after LCAM removed 50% of eligible projection-matrix weights, versus 89.5% dense and 80.5% for the strongest baseline. This is evidence only for this setting, not efficiency gains.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.