Reproduce “PRISM: Projection-Integrated Sampling-Based MPC with Bayesian Cost Tuning for Bimanual 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
Alinjar Dan, Iryna Hurova, Karl Kruusamäe, Arun Kumar Singh
PRISM, a model-based control framework for bimanual manipulation, leverages a GPU-accelerated physics simulator, projection-integrated sampling, and Bayesian cost tuning to achieve robust real-time planning and execution in contact-rich, cluttered environments, demonstrating high success rates and efficient sim-to-real transfer.
Key insights
- Decoupling exploration from trajectory smoothness and kinematic feasibility via QP projection enables robust sampling over a wider action space, leading to higher-quality motion plans for complex tasks.
- Automated offline tuning of cost weights using Bayesian Optimization significantly enhances planner robustness and success rates, alleviating manual effort and improving generalization across task instances.
- Online model-based planning with efficient physics simulation can adapt to unforeseen geometric changes and obstacles more effectively than some learned policies, without requiring retraining.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.