Flexible online planning based residual space object de-spinning for dual-arm space-borne maintenance

Abstract

De-spinning a Residual Space Object is an important premise for space-borne maintenance, especially when the relative pose between a chaser and a spinning object is inaccurate due to measurement errors. In this paper, an online trajectory planning strategy for Residual Space Object de-spinning mission under the dual-arm space robotic set-up is proposed. A novel method named Soft Recurrent Actor-Critic is formulated to plan a flexible dual-arm approaching strategy, wherein the minimum requirement of sensors are only two on-board cameras. The Recurrent Neural Network is introduced to cope with the measurement errors. Moreover, the collision limitation and the minimal joints torque variant constraint are taken into account in reward function design for safety concerns. The effectiveness of the proposed method is demonstrated in two simulations, results show that nearly 30 percent performance improvement over the two state-of-the-art methods was achieved.