GRaCE: Balancing Multiple Criteria to Achieve Stable, Collision-Free, and Functional Grasps

TL;DR

The paper tackles multi-criteria 6-DoF grasping by introducing GRaCE, a probabilistic, rank-preserving utility framework that can trade off conflicting criteria via a hierarchical rule set. A key contribution is GRaCE-OPT, a hybrid gradient-free/gradient-based optimizer that maximizes the expected grasp utility under uncertainty, leveraging a lower-bound surrogate for gradient steps. The approach is validated in both simulated (Shelf and Diner in IsaacGym) and real-world (Panda robot) experiments, showing improved sample efficiency and higher success rates compared with a traditional sample-and-filter baseline. The modular design enables on-the-fly inclusion or removal of criteria and demonstrates the practical viability of multi-criteria, intention-aware grasping in cluttered environments. These results highlight GRaCE’s potential for flexible, robust robotic manipulation in real-world tasks where multiple objectives must be balanced.

Abstract

This paper addresses the multi-faceted problem of robot grasping, where multiple criteria may conflict and differ in importance. We introduce a probabilistic framework, Grasp Ranking and Criteria Evaluation (GRaCE), which employs hierarchical rule-based logic and a rank-preserving utility function for grasps based on various criteria such as stability, kinematic constraints, and goal-oriented functionalities. GRaCE's probabilistic nature means the framework handles uncertainty in a principled manner, i.e., the method is able to leverage the probability that a given criteria is satisfied. Additionally, we propose GRaCE-OPT, a hybrid optimization strategy that combines gradient-based and gradient-free methods to effectively navigate the complex, non-convex utility function. Experimental results in both simulated and real-world scenarios show that GRaCE requires fewer samples to achieve comparable or superior performance relative to existing methods. The modular architecture of GRaCE allows for easy customization and adaptation to specific application needs.

Figures (12)

• Figure 1: In this work, we formalize optimization of grasps under multiple ranked criteria. Our probabilistic framework, GRaCE, defines an expected grasp utility $U$ where blue regions indicates higher utility values that are collision free and stable. We present a hybrid optimization method (GRaCE-OPT) for finding grasps that maximize $U$.
• Figure 2: Shelf and Diner benchmark environments with sample grasps (in blue) of high utility.
• Figure 3: Convex decomposition of the gripper used for Collision Detection Classifier.
• Figure 4: Results on Experiments on the Shelf (top) and Diner (bottom) Environments. The bar graphs show averages with standard deviation as error-bars. Using 50 samples, GRaCE outperforms Filter (5000 samples) and takes less computational time.
• Figure 5: Selected objects for intention evaluation: a fork, pan, scissors, and spatula.

Theorems & Definitions (1)

• Definition 1