A Method for Redirecting Humanoid Robots Based on Segmented Geometric Inverse Kinematics

Abstract

The current vision-based dexterous teleoperation system for robotic upper limbs mainly focuses on the recognition and reconstruction of human hand and finger postures, while the consideration of the overall arm postures is relatively insufficient, resulting in a non-negligible accumulation of errors in specific operation scenarios. To address this problem, this study proposes a novel teleoperation framework based on the human skeleton tree topology. The method captures the operator's complete upper limb posture information in real time through optical sensing devices, maps the human body motion to the robot joint space using an optimised skeleton reorientation algorithm, and achieves closed-loop feedback through a stereo vision system. The experimental results show that the proposed method has significant improvement in motion acquisition efficiency, mapping accuracy and system robustness, and provides reliable technical support for accurate teleoperation in complex environments. This study provides a new research idea and implementation way for human-robot collaboration in robot upper limb teleoperation.