High-level robotics motion planning requires the robot to accomplish complex tasks instead of simple go-to-goal navigations. This seminar presents a research topic of motion planning in robotics via the integration of reinforcement learning (RL) and linear temporal logics (LTL). The framework consists of a Markov decision process (MDP), which approximates the agent(robot)-environment interactions, and an LTL-induced automaton that expresses the user-specified task. Reinforcement learning, one of the machine learning methods, is applied to the product of the MDP and the automaton to achieve the optimal policy for the robot to complete the task. The developed framework can handle infeasible tasks, which the robot will accomplish as much as possible.
Dr. Xiao is an associate professor in the Department of Mechanical Engineering at The University of Iowa. He was graduated from Northwestern University with a Ph.D. degree in mechanical engineering before joining The University of Iowa. In the past several years, he has devoted his efforts to artificial intelligence (AI) and its applications in engineering problem-solving. His group's research interests include finite element methods, molecular dynamics, multiscale modeling and simulation, complex systems, robotics, and quantum computing.