Reproduce “GraspQP: Differentiable Optimization of Force Closure for Diverse and Robust Dexterous Grasping” 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
René Zurbrügg, Andrei Cramariuc, Marco Hutter
GraspQP generates dexterous-hand grasps using a differentiable, friction-aware force-closure objective in a constrained quadratic program and MALA*, which adapts search across candidate proposals. It releases a dataset based on 5,700 DexGraspNet objects, spanning five grippers and three grasp types. In simulation, its 12-contact power-grasp setting reaches 52% UGR versus 45% for prior methods, but does not demonstrate real-robot performance.
Key insights
- An implicitly defined Quadratic Program, differentiated via KKT conditions, provides a robust and physically accurate force closure metric for gradient-based grasp optimization that includes friction.
- Mitigating mode collapse through distribution-aware dynamic resetting and adaptive temperature scaling in the MALA* optimizer is essential for discovering a wide range of unique and stable grasp configurations.
- Maintaining hard constraints within the force closure QP formulation is critical for physical realism and performance, as simplified or unconstrained formulations lead to significant degradation in grasp quality and diversity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.