A Haptic Exploration-Inspired Control Strategy for Adaptive Grasping of Deformable Objects

Abstract

Grasping strategies based on preset thresholds or end-to-end learning often struggle to generalize to unfamiliar objects with diverse shapes, compliance, weights, or surface properties. Adaptive grasping of unstructured objects can be framed as a trade-off between excessive squeezing and unintended slippage, where accurately comprehending and effectively utilizing haptic tactile information plays a critical role. To address this challenge, we introduce a proactive preload phase that explores the object's tactile attributes. By combining the displacement of the manipulator during pinching with the measured contact force, we estimate the compliance of the object, which is then used to determine an appropriate upper bound for the grasping force. By leveraging the direct correlation between tangential contact force and the degree of object slippage, our control strategy enables real-time adaptive adjustment of the minimum required grasping force. The proposed method achieves a slippage detection accuracy of 94.21% and remains effective even for deformable heavy objects, such as a water-filled plastic bottle weighing over 2 kg. The entire process operates without the need for prior knowledge of the object or complex neural network computations. Extensive experimental validation demonstrates its robustness and generalizability across various manipulators and tactile sensors.