A Hybrid Optimization Framework for Grasp Synthesis under Partial Observations

2026-06-16

AI Summary

This work addresses the challenge of insufficient robustness in grasp pose generation under partial point cloud observations by proposing a novel approach that integrates data-driven energy priors with geometric optimization. The method uniquely combines a learned energy-based model (EBM) and the Iterative Closest Point (ICP) algorithm within a Stein Variational Gradient Descent (SVGD) framework, enabling energy-guided iterative refinement for efficient grasping of unseen objects. Experimental results demonstrate that the proposed approach achieves a success rate of 60.9% across 5,360 grasp attempts on 67 objects, significantly outperforming state-of-the-art methods including AnyGrasp (31.1%), GPD (48.4%), and AS-ICP (56.6%), thereby confirming its superior generalization capability under partial observability.

Abstract

We propose a hybrid grasp synthesis framework that combines a learning-based Energy-Based Model (EBM) with an analytical Iterative Closest Point (ICP) method to generate robust grasps from partially observed point clouds. The learned energy function acts as a prior within a Stein Variational Gradient Descent (SVGD) framework, guiding iterative refinement of grasp configurations. Evaluated on 67 objects with 5,360 grasp attempts, our method achieves an average success rate of 60.9\%, outperforming AnyGrasp (31.1\%) and Grasp Pose Detection (48.4\%) and AS-ICP (56.6\%). These results highlight the strong generalization ability of our approach and demonstrate how combining data-driven learning with geometric optimization addresses the limitations of either strategy in isolation.

Problem

• grasp synthesis
• partial observations
• point clouds
• robust grasping
• object manipulation

Innovation

• Energy-Based Model
• Iterative Closest Point
• Stein Variational Gradient Descent
• Grasp Synthesis
• Partial Observations