SpringGrasp: Synthesizing Compliant, Dexterous Grasps under Shape Uncertainty (2404.13532v2)

Abstract: Generating stable and robust grasps on arbitrary objects is critical for dexterous robotic hands, marking a significant step towards advanced dexterous manipulation. Previous studies have mostly focused on improving differentiable grasping metrics with the assumption of precisely known object geometry. However, shape uncertainty is ubiquitous due to noisy and partial shape observations, which introduce challenges in grasp planning. We propose, SpringGrasp planner, a planner that considers uncertain observations of the object surface for synthesizing compliant dexterous grasps. A compliant dexterous grasp could minimize the effect of unexpected contact with the object, leading to more stable grasp with shape-uncertain objects. We introduce an analytical and differentiable metric, SpringGrasp metric, that evaluates the dynamic behavior of the entire compliant grasping process. Planning with SpringGrasp planner, our method achieves a grasp success rate of 89% from two viewpoints and 84% from a single viewpoints in experiment with a real robot on 14 common objects. Compared with a force-closure based planner, our method achieves at least 18% higher grasp success rate.

• The paper introduces the SpringGrasp metric, a novel analytical framework for dynamic and compliant grasp synthesis under shape uncertainty.
• It leverages a GPIS-based probabilistic model to optimize pre-grasp poses and per-finger impedance control, outperforming traditional force-closure methods.
• Experimental validation on 14 objects from multiple viewpoints demonstrates robust performance and practical potential in real-world robotic manipulation.

SpringGrasp: Synthesizing Compliant, Dexterous Grasps under Shape Uncertainty

Introduction

The paper introduces SpringGrasp, a grasp planning method targeting robust and compliant dexterous manipulations in presence of shape uncertainty, common in practical robot perception due to noisy sensory inputs. A novel analytical and differentiable grasp metric, the SpringGrasp metric, is central to this approach. It evaluates dynamic grasping behaviors, enabling grasp synthesis that accounts for uncertain object surfaces and optimizes both pre-grasp hand poses and per-finger impedance control.

Problem Statement and Prior Work

Conventional robotic grasping strategies often rely on precise object models, which are not feasible in real-world scenarios involving occlusions and sensor noise. Previous solutions include data-driven approaches that learn from grasp examples and optimization-based methods focusing on stable contact point computation. However, both approaches fall short in dynamic, real-world situations where object models are uncertain and incomplete. SpringGrasp addresses this by introducing compliant grasp synthesis that dynamically adapts to the object's estimated shape and position during the grasping process.

Methodology

Compliant Grasp Formulation

SpringGrasp models a grasp not as a static pose but as a dynamic process. Fingertips equipped with impedance control move toward designated target positions, reacting compliantly to physical contact dynamics. The grasp's stability is evaluated through the introduced SpringGrasp metric, which ensures that the contact forces at equilibrium satisfy force and torque balance and remain within friction limits.

Grasp Planning with Uncertain Object Surfaces

The process begins with a GPIS (Gaussian Process Implicit Surface) model, which generates a probabilistic representation of the object's surface from noisy sensor data. The grasp synthesis optimizes for a pre-grasp pose that maximizes the certainty and efficacy of subsequent contacts. The optimization function integrates multiple objectives, including compliance, contact force estimation, target achievement inside the object geometry, and collision avoidance.

Experimental Setup and Results

The method was validated with a real robot setup handling 14 common objects, achieving superior performance against a force-closure-based baseline. Specifically, SpringGrasp demonstrated an 89% grasp success rate from two viewpoints and 84% from a single viewpoint—significant improvements over existing methods.

Discussion

The experiment highlighted the feasibility and robustness of the proposed method, significantly outperforming traditional approaches in scenarios typical in everyday robotic operations. The ability to plan and execute compliant grasps under substantial shape uncertainty without explicit tactile feedback marks a significant advancement.

Conclusions and Future Work

SpringGrasp introduces a transformative approach to robotic grasp planning that effectively handles the common real-world challenge of incomplete and noisy object data. Future enhancements could include integrating more detailed dynamic models involving object mass and gravity, improving grasp prediction for objects with challenging geometries like thin walls, and training deep learning models on datasets generated by the planned approach to predict compliant grasps from partial observations directly.

In conclusion, the proposed method not only advances the field of robotic manipulation in practical, uncertain environments but also opens avenues for further research and application in more complex and dynamic interaction scenarios.