Real-time Trajectory Planning for Redundant 7-dof Space Manipulators with Unknown Link Lengths

Trajectory planning for redundant space manipulators in complex spatial environments is critical for ensuring successful on-orbit task execution, especially when the manipulator’s link lengths are unknown. Traditional algorithms struggle to establish the transformation between task space and joint space under such conditions, posing significant challenges to autonomous trajectory planning. This study proposes a novel trajectory planning method for a seven-degree-of-freedom redundant space manipulator with a shoulder-wrist offset. The method utilizes a trained pose control decision system to generate continuous, smooth, and singularity-free trajectories in real time, satisfying positional and orientational accuracy requirements for target capture. A neural network with strong nonlinear fitting capabilities is employed as the position control decision module for joints 1-4, trained using a "cautious exploration, greedy exploitation" strategy for efficient and safe trajectory optimization. For joints 5-7, an analytical orientation control decision module adjusts the end-effector’s attitude. Together, these modules form a pose control decision system that adheres to constraints on joint angular velocity, acceleration, and singularity avoidance. The proposed approach is validated through simulations in both static and dynamic scenarios, with simulation results confirming its effectiveness.