基于技能匹配、学习、拓展的遥操作机器人控制研究

Telerobots play an important role in traditional application fields such as space and subsea exploration, advanced manufacturing and assembly, dangerous and hazardous objects manipulation, etc. With emerging applications in medical operation, health care, rehabilitation, domestic services and edutainment, a telerobot controller nowadays is required to own human-like dexterity and versatility, in addition to stability, robustness and high precision. To improve teleoperation efficiency and to enhance the tele-operator's user experience, it is expected that the telerobot controller can make intelligent adaptation to individual users according to their motion characteristics, so that a better match between the human operator and the robot can be achieved. Based on the thorough understanding and solid research background we have built in neuromotor control and adaptation mechanism, this project aims at developing effective and efficient methods to transfer dexterous and versatile skills from human operator to telerobots, which is viewed superior to traditional learning-by-demonstration approach. To ensure an optimal match between robot response and human operation, a personalized human motor model will be built to capture the user-specific neuromotor characteristics by employing advanced control theories and intelligent learning techniques. We will further enable the telerobot to generalize the learned skills both temporally and spaciously, producing super-human performance. The telerobot controller can then adjust to achieve optimal human-robot collaborations. We expect the matching, learning and generalization approach of telerobot control proposeed in this project will bring positive impacts on the development of next-generation telerobot, and push forward the study on human-robot collaboration.

遥操作机器人在海空探测、制造装配、抢险除害等传统领域发挥着重要作用。随着其在医疗康复、个人护理、家政服务和教育娱乐等新领域应用的不断增长，遥操作机器人控制器在稳定性、鲁棒性、可靠性和高精度等基础指标上更被要求具备可与人媲美的智能。为了增强遥操作员用户体验、提高人机合作效率，机器人控制器还需要智能调节来保证其响应与操作员个性化特征达到最优匹配。本项目将利用智能学习算法和先进控制理论对人的运动神经控制机理建模。在深刻理解神经运动元控制和自调节原理的基础上，开发超越传统模仿示教的技能学习手段，将操作员灵巧多用技能有效传递给机器人。进一步，项目研究将使机器人在时间、空间上对所学技能进行拓展，实现性能提升。为实现人机匹配，本项目拟根据不同操作员动作特征建立个性化神经控制模型，并据此对机器人控制器在线优化，实现最佳人机合作。本项目将有利于新一代遥操作机器人技术的发展，并能有力促进人机合作技术的深入研究。

遥操作机器人在海空探测、制造装配、抢险除害等领域应用广泛，这也对遥操作机器人控制器在稳定性、鲁棒性、可靠性和精度指标等方面提出了要求。为了增强遥操作员用户体验、提高人机合作效率，机器人控制器需要具有智能调节功能来保证其响应与操作员个性化特征达到最优匹配。针对机器人遥操作控制中的问题，本项目开展了以下工作：.1、本项目针对遥操作机器人控制理论与应用，围绕Baxter双臂机器人搭建了一套结合图像、声音、肌电信号以及力等多信息融合的遥操作实验平台。.2、开展了表面肌电信号人体刚度建模及信号处理研究，提出了基于sEMG包络的刚度增量模型提取方法以代替传统的绝对刚度提取方法，减小了非线性误差，缩短了人机矫正的时间。并开展了基于遥操作的远程抗干扰实验，远程示教写字实验，远程提升物体实验和远程康复实验。.3、搭建了一套人机技能传递界面，构建了基于远程阻抗控制的人机技能传递的机器人系统，平台主要包括机器人，肌电采集阵列，人机技能传递界面，力传感器等。进行了基波变量和神经网络自适应遥操作系列实验，实现人机遥操作良好的技能传递和交互，并通过对遥操作控制器的设计达到良好的控制性能，提高了遥操作的透明性。.4、为提高遥操作机器人系统在非结构化环境中的适应能力，课题组提出了一种基于肌电信号的变增益控制和震颤抑制方法，该方法能够实现基于个人肌电信号的变增益控制， 减小对操控者操控技能的依赖，同时抑制操控者的手部震颤，保证遥操作的可控性。.5、探索了远程机器人学习机制和感知信息的关系，提出了一种基于隐半马尔科夫和混合高斯模型的融合机器人学习技术和感知机制的机器人智能框架，可以减轻操作人员的压力，提升人机协作的性能。