【报告摘要】In recent years, the inverse optimal control problem has been a promising topic and has found applications in many areas such as bionics, collaborative manufacturing, and rehabilitation robots. In this talk, we first present some recent development on finite-horizon inverse linear quadratic regulator (LQR). By considering the inverse problem as an identification problem, its model structure is shown to be strictly globally identifiable, and a statistically consistent estimator is designed to reconstruct the unknown cost function from noisy observations of the expert demonstrations. Furthermore, since linear optimal filtering is dual to linear quadratic regulator, we then discuss how to solve the inverse filtering problem using a duality-based approach. Critical issues such as implicit observation equations and redundancy in parameter space are tackled. Finally, some extensions to inverse dynamic games are also discussed.

【报告人简介】中国科学院数学与系统科学研究院优秀青年副研究员。主要从事动态规划、辨识与滤波、网络系统分布式控制等领域的理论与应用研究。具体方向包括逆最优控制、动态博弈系统、多智能体系统的协同定位与编队控制等。