A Method for Constructing a Dual-Arm Robot Motion Retargeting Dataset

Fracture reduction is a delicate surgical task that intrinsically falls within the realm of human-robot interaction. The human-likeness of arm posture directly influences both the reduction efficiency and the patient's acceptance. To address the challenges of high cost and the difficulty in balancing end-effector accuracy with human-likeness during motion retargeting dataset construction, this paper presents a human–robot motion retargeting dataset construction method based on the Adam optimization algorithm. The method employs the Arm-Configuration Angle of robot arm as a key metric for quantifying human-likeness, while introducing end-effector position and orientation as precision constraints to formulate a multi-objective optimization model for accurate mapping of human arm motion data to dual-arm robotic movements. By integrating Arm-Configuration Angles and end-effector posture into a unified optimization framework, a multi-task loss function is designed, and the Adam optimizer is applied to iteratively solve for the retargeted joint angles. Simulation experiments demonstrate that the proposed method effectively maintains the motion consistency between the robot end-effector and human hand, while preserving a high degree of human-like characteristics in the joint space. Specifically, the probability of end-effector retargeting accuracy below 2mm reaches 98 %, and the Arm-Configuration Angle error in the joint space remains within 1°. Further validation on a real dual-arm robotic platform shows that the generated motion data not only fulfills the precision demands of fracture reduction but also significantly enhances the naturalness and generalizability of retargeted actions.