Intelligent Haptic Controls for Robotic Teleoperation

计算机任务自动化通常实现“控制系统”，就像汽车的巡航控制系统一样，通过控制油门来保持速度恒定。一个简单的反馈控制通过将输出误差乘以一个负常数来决定控制输入；在巡航控制的情况下，油门控制与速度误差成正比。更精确的反馈控制律通常是通过对误差信号进行更复杂的数学运算，特别是积分和微分来实现的。另一方面，我的研究重点是通过生物学启发的方法来改进控制系统，特别是利用小脑模型算术计算机（CMAC），这是一种以人脑小脑为模型的计算机算法。这种“人工神经网络”将学习和记忆方面纳入控制，从而适应未知变量（如巡航控制示例中汽车乘客的重量）。在这个提案中，我将把CMAC应用到机器人“触觉”这个令人兴奋的新领域——通过手控制器来控制机器人机械臂，让人们感觉和/或指挥力量。应用范围从Canadarm2/Dextre到潜艇操纵器再到手术机器人。大多数手控机器人系统只允许人们控制运动（位置或速度），而严格依靠他们的眼睛来知道机器人在做什么。现在，一些现代机器人系统可以让人们在控制机器人的同时感受机器人所遇到的力；触觉提供了更直观的交互，从而实现更快、更准确的操作。当前的控制系统技术采用数学反馈规律，在某些情况下，如系统中存在时滞或环境突然变化时，可能无法给用户产生满意的感觉。考虑一个外科手术机器人，理想情况下，它应该能够在病人身体外移动，用人体组织执行手术任务，即使外科医生没有感觉到与机器人完全相同的力，也能毫无困难地推动骨骼（以及在这些任务之间转换）。根据目前的技术，带有触觉的手术机器人是为一种特定类型的任务而设计的，而要完成另一种任务则需要很大的困难，或者根本不需要。可能遇到的特殊问题包括急剧的突然运动、不必要的振动、刺穿时的过度运动以及碰撞时的过度作用力。我将使用CMAC来改进控制系统，使机器人能够自然地适应不同环境的运动/力，同时反应更像人。通过更多地依赖CMAC的记忆能力，而不仅仅是误差测量，运动可以变得更平滑，振动更少，甚至没有。在两个不同的CMAC控制之间平滑切换也成为可能，因此在穿刺或碰撞时，可以在适当的控制之间平滑过渡-避免极端的运动或力量。最重要的是，在数学框架内开发CMAC控制，以保证在不可避免的小时间延迟存在时的稳定性。因为人们将体验到系统更加直观，他们将需要较少的事先正式培训，并且在控制系统时经历较少的压力。卡尔加里山麓医院的神经手臂项目向我开放了他们的触觉实验室，这将使我有机会使用最新的触觉设备来测试和评估我的控制，并与每天使用这种系统的外科医生密切合作。

Computer automation of tasks often implements "control systems", like a car's cruise-control that keeps the speed constant by controlling the throttle. A simple feedback control decides on the control input by multiplying the output error by a negative constant; in the case of the cruise control the throttle control is proportional to the error in speed. More accurate feedback control laws are normally achieved by performing more complex mathematical operations on the error signal, especially integration and differentiation. On the other hand, my research focuses on improving control systems through biologically inspired methods, specifically by utilizing the Cerebellar Model Arithmetic Computer or CMAC, a computer algorithm modelled on the cerebellum in the human brain. Such "artificial neural networks" incorporate a learning and memorization aspect into the control, resulting in adaptation to unknown variables (like the weight of the passengers in the car in the cruise control example). In this proposal, I will apply the CMAC to the new and exciting area of robotic "haptics" - the control of robotic manipulator arms through hand controllers that let people feel and/or command the forces. Applications range from the Canadarm2/Dextre to submarine manipulators to surgical robots. Most hand-controlled robotic systems only let people control movement (position or speed) while relying strictly on their eyes to know what the robot is doing. Some contemporary robotic systems now let people control the robot while also feeling the forces encountered by the robot; the sense of touch provides a more intuitive interaction resulting in faster and more accurate manipulation. Current control system technology, using mathematical feedback laws, may not produce a satisfactory feel to the user under certain conditions, like when there are time-delays in the system or the environment changes suddenly. Consider a surgical robot, which ideally should be able to move outside the patient, perform surgical tasks with human tissue, and push against bone (and transition between these tasks) without difficulty even when the surgeon does not feel the forces at exactly the same time as the robot. With current technology, surgical robots with haptics are designed for one particular type of task and can do another only with great difficulty, or not at all. Particular problems that can be encountered include sharp abrupt movements, unwanted vibrations, excessive movement during punctures, and excessive forces during collisions. I will use the CMAC to improve the control systems, so that the robot will naturally adapt its motions/forces to different environments while, at the same time, reacting more like a person would. By relying more on the CMAC's memorization capabilities rather than only on measurements of error, the motions can be made much smoother with less, or no, vibrations. It also becomes possible to smoothly switch between two different CMAC controls, so that during puncture or collision a smooth transition can be made between appropriate controls - avoiding extreme movement or forces. It is of utmost importance that the CMAC controls be developed within a mathematical framework that guarantees stability in the presence of the small unavoidable time delays. Because people will experience the system as more intuitive, they will require less formal training beforehand and experience less stress when controlling the system. Project NeuroArm at Foothills Hospital in Calgary has opened their haptics lab to me, which will give me the opportunity for testing and evaluating my controls using the latest in haptic equipment in close consultation with surgeons who use such systems on a daily basis.