Reproduce “ActivePusher: Active Learning and Planning with Residual Physics for Nonprehensile 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
Zhuoyun Zhong, Seyedali Golestaneh, Constantinos Chamzas
ACTIVEPUSHER introduces a framework for robotic nonprehensile manipulation that integrates residual physics modeling, Neural Tangent Kernel-based uncertainty quantification, and active learning with kinodynamic planning. This approach enables data-efficient learning of dynamics models and generates more robust plans, achieving high task success rates in real-world pushing tasks with significantly less interaction data than prior methods.
Key insights
- Combining residual physics with uncertainty-based active learning significantly improves data efficiency for learning object dynamics, requiring less real-world interaction data.
- The Neural Tangent Kernel (NTK) provides an effective and computationally efficient method for quantifying epistemic uncertainty in finite-width neural networks, which can be leveraged for both active data acquisition and robust planning.
- Incorporating model uncertainty into kinodynamic planning by prioritizing low-uncertainty actions leads to more reliable and robust task execution, even if it results in slightly longer paths.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.