In order to extend promising robot applications in human daily lives, robots need to perform dextrous manipulation tasks, particularly for a mobile dual-arm robot. This paper propose a novel control strategy, which consists of a first trial process and a learning phase, to enable a mobile dual-arm robot to complete a grasp-and-place task which can be decomposed into movement sequences, such as reaching, grasping, and cooperative manipulation of a grasped object. Under the guidance of vision system, the robot with physical constraints successfully fulfills the task by tracking trajectories generated by redundancy resolution online using a neural-dynamic optimization. Then a reinforcement learning algorithm called the policy improvement with path integrals for sequences of dynamic movement primitives is applied to learn and adjust the recorded trajectories. Experimental results of the developed mobile dual-arm robot verified that the proposed strategy is able to successfully and optimally complete a grasp-and-place task.

中文翻译：

---

使用动态运动原语序列的移动双臂机器人协同操作

为了在人类日常生活中扩展有前景的机器人应用，机器人需要执行灵巧的操作任务，特别是对于移动双臂机器人。本文提出了一种新的控制策略，包括第一次试验过程和学习阶段，使移动双臂机器人完成抓取和放置任务，该任务可以分解为运动序列，如到达、抓取，以及对被抓物体的协作操作。在视觉系统的引导下，具有物理约束的机器人通过使用神经动态优化在线跟踪冗余解析生成的轨迹，成功地完成了任务。然后，使用称为动态运动原语序列的路径积分的策略改进的强化学习算法来学习和调整记录的轨迹。开发的移动双臂机器人的实验结果验证了所提出的策略能够成功并最佳地完成抓取和放置任务。