| 摘要: |
| 在空间机器人双臂协同装配任务中，机械臂与装配组件之间的接触状态呈现复杂的非线性，现有的控制技术未能解决装配过程中产生过大动态接触扰动的问题，进而导致装配精度下降、任务失败甚至设备损坏。为解决此问题，提出了一种基于深度确定性策略梯度（DDPG）算法的协同控制方法，构建了适应复杂接触状态变化的双臂协同机构，推导了包含机械臂与装配组件的动力学模型，进行了空间双臂机器人在快速、中速、慢速对接工况下的动力学分析，采用DDPG强化学习算法，降低机器人在对接过程中的接触力。通过仿真与实验验证，该方法可以显著提升装配精度（提高20%）、装配成功率（达到80.25%）、降低瞬间接触力大小（降低30%），有效调节空间双臂机器人协同装配过程的稳定性与可靠性。 |
| 关键词: 强化学习 空间机器人 双臂协同控制 触碰撞模型 机器人动力学 |
| 基金项目:国家自然科学基金项目（52475009；52175082）。 |
|
| Variable Stiffness Compliant Control Technology for Space Dual-arm Robots Based on DDPG |
|
Wang Siyu1, Niu Hairui2, Wang Kai2, Gu Haiyu*3, You Bindi1, Zhai Fujun2, Tang Jiarui2
|
|
1. School of Ocean Engineering, Harbin Institute of Technology (Weihai), Weihai 264209;2. Beijing Spacecraft Manufacturing Co., Ltd., Beijing 100094;3. School of Astronautics, Harbin Institute of Technology, Harbin 150001
|
| Abstract: |
| In space robotic dual-arm cooperative assembly tasks, there exists a complex nonlinear contact state between the robotic arms and the assembly components. Existing control techniques have not successfully addressed the issue of excessive dynamic contact disturbances during the assembly process, leading to reduced assembly accuracy, task failure, and even equipment damage. To resolve this issue, this paper proposes a cooperative control method based on the Deep Deterministic Policy Gradient (DDPG) algorithm. A dual-arm cooperative mechanism adaptable to complex contact state changes is constructed, and the dynamic model of the robotic arms and assembly components is derived. The paper conducts a dynamic analysis of the space dual-arm robot under rapid, medium, and slow docking conditions. The DDPG reinforcement learning algorithm is applied to reduce the contact force during the docking process. Through simulation and experimental validation, the proposed method significantly enhances assembly accuracy (by 20%), assembly success rate (reaching 80.25%), and reduces instantaneous contact force (by 30%). The approach effectively regulates the stability and reliability of the space dual-arm robot cooperative assembly process. |
| Key words: reinforcement learning space robot dual-arm cooperative control contact collision model robot dynamics |
