Zhang Shiqi

Associate Professor, Binghamton, State University of New York

Leveraging Large LanguageModels for Robot Planning inOpen Worlds

Report time: 10:00, June 19, 2023

Venue :61#5123

Research areas: Artificial intelligence, robotics, human-robot interaction and cooperation

Organizer: College of Intelligent Science and Engineering

Brief introduction of reporter

Dr. Shigi Zhang is an Associate Professor with the Department of ComputeScience, the State University of New York (SUNY) at Binghamton. Before joiningSUNY Binghamton, he was an Assistant Professor at Cleveland State Universitafter working as a Postdoc at UT Austin. He received his Ph.D. in ComputtScience (2013) from Texas Tech University, and received his M.S. and B.Sdegrees from Harbin institute of Technology. He is leading an NSF NationtRobotics lnitiative project on robot decision making. He received the Best Robotic!Paper Award from the AAMAS conference in 2018, a Ford URP Award in 2019.and an OPPO Facutv Research Award in 2020.

Brief introduction to the report

Robots need task planning algorithms for sequencing high-level actions,and motion planning algorithms for realizing those task-level action. Task and motion planning(TAMP) algorithms are for interleaving those two types of planning paradigms to ensure task completions. The real world is generally "open" with new objects and newt situation, where classical planning methods do not perform well. Recent advances in pretrained large language models (LLMS) have reshaped the landscape of AI. In this talk,I'll present our recent efforts on robot TAMP with and without LLMs to facilate its task completions with unforeseen situations. We have used a mobile manipulator working on everyday service tasks for evaluation and demonstration purposes.