Towards A Reliable Optimization-based Design Framework for Autonomy and Control of Robotic Systems

自主机器人车辆系统（ARVs），如无人驾驶飞行器、自主水下航行器和移动机器人，提供了有前途的工具，使人们从无聊、重复或危险的工作中解脱出来，以高效、自主和经济的方式完成各种有意义的任务。ARV实际应用场景的复杂动态环境要求对ARV进行准确、可靠的控制。然而，基于传统线性控制理论的自主控制系统往往带有很强的假设，这些假设在实践中可能不被满足，从而导致性能不佳。处理系统约束（如有限的传感、计算和执行能力）的困难进一步排除了使用常规方法实现最佳性能的可能性，因为最佳性能很可能位于约束有效的边界上。从技术上讲，控制设计问题可以表述为优化问题。通过优化设置，有可能克服这些限制并发展下一代抗逆转录病毒技术。提出的研究方案旨在将先进的优化技术整合到ARV控制和自治层设计中，并解决最具挑战性的问题，包括：(1)可靠性：由于优化求解器可能无法给出最优解，我们如何保证开发的基于优化的控制系统不受此类故障的影响并实现指定的控制目标？(2)适用性：由于控制过程中存在不确定性/干扰，我们如何量化期望性能与不确定性水平之间的关系，以此来判断控制设计在规定不确定性的情况下是否能满足性能要求？(3)可扩展性：开发的框架应该能够指导控制和自治层的设计，不仅适用于单个ARV，也适用于一个团队，那么我们如何证明可扩展性？(4)实时控制：通过迭代方法求解优化问题，可能需要相当长的时间才能收敛到一个解。由于机器人系统是快速动态系统，因此需要在不到十分之一秒的时间内获得解决方案。那么我们如何设计处理管道和/或特别快速优化算法来满足实时控制要求呢？该方案将回答上述重要问题，并为ARV控制系统设计开发一种新颖的分析、综合和设计工具包；它将提高自主机器人系统的可靠性和可操作性，降低其运行过程中的风险和成本。此外，该项目将使创新和智能机器人应用受益于加拿大社会，并为加拿大快速增长的机器人行业提供培训HQP的巨大机会。

The autonomous robotic vehicle systems (ARVs), such as unmanned aerial vehicles, autonomous underwater vehicles and mobile robots, present promising tools to release people from boring, repetitive, or dangerous jobs, and accomplish various meaningful tasks in an efficient, autonomous, and cost-effective way. The complex and dynamic environment in practical ARV application scenarios requires accurate and reliable control of the ARV. However, the autonomous control system based on conventional linear control theories often come with strong assumptions which may not be satisfied in practice and hence lead to poor performances. The difficulty in handling system constraints (such as limited sensing, computing and actuating capabilities) further rules out the possibility of using conventional methods to achieve optimal performance as the optimum is likely to be located on the boundary where the constraints are active. Technically, the control design problems can be formulated as optimization problems. With the optimization setup, it is possible to overcome these limitations and to evolve next-generation ARV technologies. The proposed research program aims to integrate advanced optimization technology in the ARV control and autonomy layer design and address the most challenging issues including: (1) Reliability: Since the optimization solver may fail to give an optimal solution, how do we guarantee the developed optimization-based control system will not be affected by such failures and will achieve the designated control goal? (2) Applicability: Since there exist uncertainties/disturbances in the control process, how do we quantify the relationship between expected performance and the level of uncertainty, based on which we judge whether the control design will meet performance requirement in presence of specified uncertainty? (3) Scalability: The developed framework should be able to guide the control and autonomy layer design not only for a single ARV but also for a team of them, so how can we justify the scalability? (4) Real-Time Control: The optimizations are solved by iterative methods which may take considerable time to converge to a solution. Since robotic systems are fast dynamic systems, a solution needs to be obtained within less than tenth of a second. So how do we design the processing pipeline and/or ad hoc fast optimization algorithms to meet the real-time control requirement? The proposed program will answer the above important questions and develop a novel analysis, synthesis and design toolkit for the ARV control system design; it will improve the reliability and operability of autonomous robotic systems and lower the risks and costs during their operations. Furthermore, this program will benefit the Canadian society by enabling innovative and intelligent robotic applications and provide tremendous opportunities for training HQP for the fast-growing robotics industry in Canada.