讲座名称：基于学习的最优控制和强化学习研究

讲座人：高伟男 副教授

讲座时间：9月10日10:00

讲座地点：腾讯会议直播（ID：618 948 582 链接https://meeting.tencent.com/dm/6tadVLNeKi7W）

主持人：平续斌

讲座人介绍：

高伟男，2011年获得东北大学自动化专业学士学位；2013年获得东北大学控制理论与控制工程硕士学位；2017 年在纽约布鲁克林的纽约大学获得电气工程博士学位。在2017-2020年期间，任乔治亚州艾伦 E.保尔森工程与计算集团电气与计算机工程助理教授，以及佐治亚州斯泰茨伯勒南方大学助理教授；并且在2018年在马萨诸塞州剑桥市三菱电机研究实验室做访问教授。自2020年起担任佛罗里达州佛罗里达理工集团的工程与科学集团机械与土木工程系副教授。他的研究兴趣包括强化学习、自适应动态规划、最优控制、协同自适应巡航控制、智能交通系统、采样数据控制系统和输出调节理论。2018年获得IEEE实时计算与机器人国际会议最佳论文奖，2019年获得纽约大学David Goodman研究奖。目前现任IEEE/ CAA Journal of Automatica Sinica，Neurocomputing，Control Engineering Practice编委员会成员，Complex & Intelligent Systems 客座编辑，Neural Computing and Applications 编委会成员，IEEE非线性系统和控制以及 IFAC TC 1.2 自适应和学习系统等控制系统学会技术委员会成员。

报告1: Learning-Based Adaptive Optimal Control and Applications to Connected and Autonomous Vehicles

摘要: The connected and autonomous vehicle (CAV) technology can prevent secondary crashes, reducing property damage and injury, congestion, and emissions. Among all CAV studies, the controller design of CAV has attracted considerable attention among researchers in the field of control, optimization, and communication. In this talk, I will introduce several intelligent cruise control design strategies under the framework of reinforcement learning andadaptive dynamic programming to address the adaptive optimal control problem of CAVs. Approximate optimal control policies are learned by collected online data from vehicles without relying on the knowledge of either human or vehicle models. Microscopic traffic simulation results show that these approaches can increase traffic throughput while reducing fuel usage.

报告2：Reinforcement Learning and Adaptive Dynamic Programming

摘要：Reinforcement learning (RL) concerns how an agent interacts with unknown environment in order to maximum the cumulative reward. As a branch of RL, Adaptive Dynamic Programming (ADP) is a sound data-driven, model-free approach for optimal control design of complex dynamic systems. Output regulation is a general mathematical framework that designing controllers to achieve disturbance rejection and asymptotic tracking of a dynamic system. I will introduce a novel framework of adaptive optimal output regulation via RL and ADP. Under this framework, one can design adaptive optimal controllers to solve output regulation of linear systems, nonlinear systems, and multi-agent systems.

主办单位：机电工程集团