Formal Behavioral Specification Languages and Methods for Motion Planning and Control of Autonomous Systems
Thu. Oct. 26th from 10am - 12:00pm EST
Speaker: Yash Vardhan Pant - Assistant Professor in the Department of Electrical and Computer Engineering at the University of Waterloo
YouTube Link to this video here.
CPI Talks are free and open to everyone regardless of affiliation! High school students and non-Waterloo students/staff are also welcome to join.
No prior knowledge will be expected from the audience.
In this CPI Talk, Yash Vardhan Pant discussed:
Formal Behavioral Specification Languages and Methods for Motion Planning and Control of Autonomous Systems
Abstract:
Safe planning and control of single and multi-agent autonomous systems performing complex tasks has been a challenging problem. Recently, reinforcement learning-based solutions have shown great promise towards achieving these goals. However, they (usually) suffer from a lack of strong guarantees on safety, and potentially, the problem of reward hacking. In this talk, I will instead look at this problem of behavioral planning through the lens of formal behavioral specification languages, namely Signal Temporal Logic (STL) and Linear Temporal Logic (LTL), which offer a sound, succinct and unambiguous way of representing complex desired behaviors for autonomous systems. I will present a family of robust and predictive motion planning and control methods for autonomous systems with objectives defined using STL and LTL. For STL-based motion planning, we formulate efficient optimization-based methods that provide strong guarantees on performance and safety even when operating in partially known environments. For LTL-based motion planning, we present ongoing work on a model-free reinforcement learning approach that is inspired by the paradigm of self-play. The performance and scalability of the methods will be demonstrated through multi-robot simulation studies and experiments on actual quadrotor drones.
Dr. Yash Vardhan Pant is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Waterloo, where he leads the Control, Learning and Logic (CL2) group. He received his PhD in Electrical Engineering from the University of Pennsylvania in 2019 and was a postdoctoral fellow at the University of California at Berkeley from 2019-2021, before joining Waterloo in the summer of 2021. His research focuses on decision-making for multi-agent and autonomous systems, drawing on elements of Control Theory, Machine Learning, Formal Methods and Optimization, with application to ground robots, human-robot interaction and swarms of aerial robots. More details about Dr. Pant’s research can be found at: https://yashpant.github.io/