Data-driven distributed control of mobile robotic networks: Where machine learning meets game theory

移动的机器人网络;（例如，无人驾驶飞行器机队）为公认的军事用途以及各种各样的民用用途提供了扩大的能力。有几个因素促使它们的潜力和重要性日益增加。特别是，技术进步使更小的平台具有更高的传感，通信和处理能力。此外，自主操作提供了几个竞争优势，例如超过人类疲劳极限的持续监视或远程操作能力，而无需资产和人员的后勤运输成本。智能优势：分布式控制成为充分发挥移动的机器人网络潜力的关键。当前的分布式控制范例主要是基于模型的，并且不足以处理显著的不确定性，包括（1）环境不确定性;即，移动的机器人操作的非结构化环境中不可预见的元素;（2）动态不确定性;即，移动的机器人的物理动力学的不准确性。为了弥合差距，该项目将利用强化学习（机器学习的一个领域）和博弈论（最初在经济学中发展）来开发一个新的数据驱动（更具体地说，无模型）分布式控制框架。开发的框架是无模型的，完全分布式的，自治的，其性能是严格可证明的。该框架将显着提高自主性的移动的机器人，当他们面临重大的环境不确定性和动态不确定性，特别是在长期任务。更广泛的影响：这项研究的成功完成将提供工程准则的分析，合成和原型的移动的机器人网络，可以有效地在非结构化环境中运行。研究结果深刻影响了各种工程学科，包括科学数据收集，国土安全行动和智能交通系统。拟议的研究是跨学科的，涉及博弈论，机器学习，控制，机器人运动规划和分布式算法之间的相互作用。这将为STEM领域的高中、本科和研究生提供跨越传统学科界限的教育和培训机会。与工业合作伙伴的合作强调了在学术界之外产生影响的潜力。

Mobile robotic networks; (e.g., fleets of unmanned aerial vehicles) offer expanded capabilities for recognized military uses as well as a wide variety of civilian uses. There are several factors that contribute to their increasing potential and importance. In particular, technological advances have enabled smaller platforms with increased sensing, communication, and processing capabilities. In addition, autonomous operations offer several competitive advantages such as persistent surveillance that exceeds human fatigue limitations or remote operation capabilities without the logistical transport costs for assets and personnel. Intellectual Merit: Distributed control becomes key to fully realize the potentials of mobile robotic networks. Current distributed control paradigms are mainly model-based and inadequate to handle significant uncertainties, including (1) environmental uncertainties; i.e., unforeseeable elements in unstructured environments where mobile robots operate; (2) dynamic uncertainties; i.e., inaccuracies of the physical dynamics of mobile robots. To bridge the gaps, this project will leverage reinforcement learning, an area of machine learning, and game theory, initially developed in economics, to develop a new data-driven (more specifically, model-free) distributed control framework. The developed framework is model-free, fully distributed, autonomous, and its performance is rigorously provable. The framework will significantly improve the autonomy of mobile robots when they face significant environmental uncertainties and dynamic uncertainties especially in long-term missions. Broader Impacts: Successful completion of this research will provide engineering guidelines in analysis, synthesis and prototyping of mobile robotic networks which can effectively operate in unstructured environments. The research findings profoundly impact a variety of engineering disciplines, including scientific data collection, homeland security operations and intelligent transportation systems. The proposed research is interdisciplinary and involves interactions among game theory, machine learning, control, robotic motion planning and distributed algorithms. This will lead to educational and training opportunities that cross traditional disciplinary boundaries for high-school, undergraduate and graduate students in STEM. The collaborations with industrial partners stress the potentials to make an impact beyond academia.