基于模仿的仿生双臂协调作业学习方法研究

仿生双臂型机器人在工业、服务、海洋工程、外空间都有良好的应用前景，但在复杂环境下机器人如何学习自主协调双臂完成新任务仍面临巨大挑战。而通过示范、示教的方式让仿生双臂服务机器人自主学习新的复杂操作任务、适应环境变化是解决以上问题的重要研究思路。本课题在脑科学启发下，将镜像神经系统模型引入仿生双臂协调作业的示范、示教学习研究中，深入分析具有视触感知能力的仿生双臂机器人系统最优构形、镜像神经系统神经计算模型，提出人-机器人的双臂动作映射、基于模仿的仿生双臂协调作业学习框架和优化方法，并通过复杂装配任务进行验证，进而构建动作捕捉、动作映射、学习框架到应用验证的较为系统的方法体系。通过本课题的研究工作，为双臂自主学习新的复杂操作任务提供新的思路和方案，为解决自动编程、复杂操作任务自学习奠定理论和方法基础。

Robots with biomimetic dual manipulators have a brilliant prospect in the applications such as industry, service, ocean engineering, and outer space. But how to autonomously learn new coordination tasks for dual manipulators robots under complex environment is still a great challenge. For solving the above issue, an important idea is that the robots with biomimetic dual manipulators learn to fulfill new complex manipulation tasks and adapt the environment change by means of teaching or demonstration. So inspired by the brain sciences, this project tries to introduce mirror neuron system model into biomimetic dual manipulators’ coordination learning through teaching or demonstration. The optimized configuration of the biomimetic dual manipulators with visual and tactile sensing, the computational neuron model of mirror neuron system, are to be analyzed further. The motion mapping between human arms and dual manipulators, the imitation-based learning framework and optimization for coordinated manipulation of biomimetic dual manipulators, are to be presented and validated through complex assembling tasks. The systematic methods will be formed from motion capture, motion mapping, learning framework to application verification. This research would bring novel ideas and solutions to the dual manipulators’ autonomous learning for new complex manipulation tasks, and laying foundations for automatic robot programming and self-learning of complex manipulation tasks.