Development of Adaptive Learning Algorithms for High-Performance Control of Robotic Systems

9810398陈在过去的几十年里，为了提高生产率和产品质量，机器人系统的部署有了巨大的增长。这些工业机器人绝大多数用于简单的重复性任务。传统的机器人控制器没有充分利用机器人的重复性，在所有的重复执行中都会产生相同的性能误差。因此，今天超过80%的机器人被限制在要求相对较低精度的任务上。然而，据预测，未来的增长将是用于电子组装等任务的高性能机器人系统。学习控制，能够从过去的经验中学习以消除未来试验中的错误，有很大的潜力成为未来机器人系统的首选控制器。该技术的瓶颈是无法处理高速运行和具有柔性部件的机器人系统中出现的非最小相位动力学。该项目的主要目标是开发一种新的控制方法，用于具有轻型柔性部件的机器人系统精确和高速执行重复任务。技术方法是基于逆的自适应学习控制，从过去的执行数据中学习最优控制输入和系统模型。该方法将通过将稳定反演技术纳入学习算法来解决上述瓶颈，稳定反演是最近开发的用于处理非最小相位系统的理论。该项目还将通过模拟和实验测试验证新技术的有效性，并解决在实际实施中出现的技术问题，从而加快新技术在工业机器人系统中的应用。利用PI在机器人控制方面的八年经验和他在稳定反演方面的专业知识，这个为期三年的项目将提供以下内容：1)一系列自适应学习控制算法，设计用于最小和非最小相位系统；2)一套非线性工具，能够为轻型机器人系统中出现的非线性系统设计和实现自适应学习控制器；3)具有轻型柔性组件的实验试验台（可用于各种测试目的的可重构3D起重机）；4)广泛的仿真和实验测试，验证了新的控制技术，并在实际实施中展示了其高性能。除了扩展系统和控制方面的知识库之外，这种控制技术的成功交付将对涉及机器人系统的行业产生重大影响。新技术的部署将提高产品质量，因为它赋予机器人学习和消除重复错误的能力。由于新的控制器可以克服高速运行中结构灵活性带来的困难，因此生产率将更高。目前广泛使用的各种起重机，由熟练的操作员手动控制，可以使用新的控制器实现自动化，实现高速高性能，有效地将其转换为可编程机器人。该技术还促进了更轻重量机器人系统的设计和部署，从而带来更快的速度，更少的能源消耗，更少的材料浪费，以及其他环境和社会效益。＊＊＊

9810398ChenThe past few decades have witnessed tremendous growth in the deployment of robotic systems to improve production rate and product quality. The vast majority of these industrial robots are used for simple, repetitive tasks. Traditional robot controllers do not take advantage of the repetitive nature and the same performance error will persist in all repetitive executions. Consequently, over eighty percent of today's robots are restricted to tasks requiring relatively low precision. It is predicted, however, that future growth will be in high performance robotic systems for tasks such as electronic assembly. Learning Control, being able to learn from past experience to eliminate errors in future trials, has great potential to emerge as the controller of choice for future robotic systems. The bottleneck hindering this technology is its inability to handle non-minimum phase dynamics that arise in high-speed operation and in robotic systems with flexible components.The main objective of this project is to develop a new control methodology for the precise and high-speed execution of repetitive tasks for robotic systems with light-weight flexible components. The technical approach is inversion-based adaptive learning control that learns both the optimal control input and the system model from past execution data. This approach will resolve the above mentioned bottleneck by incorporating into the learning algorithms techniques from stable inversion, a theory recently developed for handling non-minimum phase systems. The project will also speed up the application of the new technology to industrial robotic systems by validating its effectiveness through simulation and experimental testing and by resolving technical issues arising in its practical implementation.Utilizing the PI's eight years of experience in robotic control and his expertise in stable inversion, the three year project will deliver the following: 1) a family of adaptive learning control algorithms designed to work for both minimum and non-minimum phase systems; 2) a set of nonlinear tools that enable the design and implementation of adaptive learning controllers for nonlinear systems that arise in light-weight robotic systems; 3) an experimental testbed with light-weight flexible components (a 3D crane reconfigurable for various testing purposes); 4) extensive simulation and experimental testing, validating the new control technology and demonstrating its high-performance in practical implementation.In addition to expanding the knowledge base in systems and control, the successful delivery of this control technology will have a significant impact on industries involved with robotic systems. The deployment of the new technology will bring about improved product quality since it empowers the robot to learn and eliminate repetitive errors. Production rate will be higher since the new controller can overcome difficulties caused by structural flexibility in high speed operation. Various cranes, currently in widespread use and manually controlled by skilled operators, can be automated using the new controller to achieve high performance at high speed, effectively converting them into programmable robots. The technology also promotes design and deployment of lighter-weight robotics systems, which in turn leads to faster speed, less energy consumption, less material waste, and other environmental and societal benefits. ***