基于深度强化学习的月面建造机器人组装挂载控制

Assembly and Mounting Control of Lunar Construction Robot Based on Deep Reinforcement Learning

月球基地建造是未来月球长期驻留科学探测与资源开发利用的必然要求,月面建造机器人的组装挂载控制是未来月面探测和月面建造中亟须突破的关键技术。为应对月面动态非结构化环境和建造任务自主性带来的挑战,规划了月面建造机器人组装挂载的任务场景,提出了基于视觉和力反馈多模感知融合的两阶段深度强化学习组装挂载控制方法,提出了基于视觉位姿范数和基于分阶段力导数的奖励函数准则。通过仿真和地面真实机器人试验验证了算法的有效性,实现了由图像和力输入到组装挂载动作直接输出的端到端控制,解决了月面建造中复杂部件的高精度机器人组装挂载问题。

The construction of lunar bases is necessary for long⁃term lunar scientific exploration and resource development and utilization in the future.The assembly and mounting control of lunar sur⁃face construction robots is a key technology that needs to be broken through in future lunar surface construction.To solve the challenges of dynamic unstructured lunar environment and the autonomous construction tasks,a two⁃stage deep reinforcement learning assembly and mounting control method based on the fusion of vision and force multi⁃modal perception was proposed,and a visual pose norm⁃based reward function and a reward function based on staged force derivatives were designed.The effectiveness of the algorithm was verified through simulation and ground real robot experiments.The method achieved end⁃to⁃end robot control from image and force input to direct output of assem⁃bly actions which could realize high⁃precision robotic assembly of complex components in lunar sur⁃face construction.