Hybrid Control Approach for Walking-Assembly Integrated Space Robots in On-Orbit Assembly*

Abstract

As space exploration advances, the demand for assembling large-scale structures in orbit, such as telescopes and space stations, continues to grow due to transportation size constraints. Robotic systems play a critical role in these tasks, requiring precise control and efficient energy management in the challenging conditions of space. This paper proposes a hybrid control strategy that combines Model Predictive Control (MPC) with Reinforcement Learning (RL) to optimize the performance of a 7-degree-of-freedom walking-assembly integrated space robot. MPC optimizes control inputs by predicting the robot's response over a defined horizon while handling constraints in real time, and RL dynamically tunes MPC parameters to adapt the control system to task-specific priorities. The proposed controller operates in two distinct modes: energy-efficient mode and accuracy-focused mode, balancing energy consumption and task precision. Simulation results validate the approach, demonstrating significant energy savings while maintaining high accuracy during space assembly tasks.