Semi-Autonomous Grasping Control of Prosthetic Hand and Wrist Based on Motion Prior Field

Xu Shi,Weichao Guo,Wei Xu,Zhiyuan Yang,Xinjun Sheng

DOI: https://doi.org/10.1109/lra.2024.3398563

IF: 5.2

2024-01-01

IEEE Robotics and Automation Letters

Abstract:Grasping multiple affordance parts and from arbitrary directions for complex shaped objects still remains a challenging problem for prosthetic hand with wrist. We propose a semi-autonomous control method that uses only an integrated in-hand camera to predict the final grasping part on an object as the hand approaches it and obtain the appropriate wrist joint angles and preshape type. We collect approach-grasp motion sequences from human experts to construct a motion prior field (MPF) and derive the prediction model MPFNet by behavior cloning. With noise augmentation and a hybrid regression-categorization policy training, our prediction model gets less than 2 cm predicting deviation under a small number (15) of demonstrations for each object. We apply our control method to a prosthetic hand with a 2 degrees -of-freedom (DoF) wrist, enabling it to grasp multiple parts of complex shaped objects and remain robust under the position and orientation variation. Compared to state-of-the-art myoelectric control and semi-autonomous control methods, respectively, our method improves 65.4%/26.3% in grasp success rate, 40.4%/26.3% in control time, and 35.6%/27.8% in error distance. Furthermore, our method is adaptable to different objects in the same category.