Motion planning under constraints and uncertainty using data and reachability

Motion planning under constraints and uncertainty using data and reachability

Speaker Name: 
Abraham Vinod
Speaker Title: 
Research Scientist
Speaker Organization: 
Mitsubishi Electric Research Laboratories (MERL)
Start Time: 
Thursday, June 8, 2023 - 2:00pm
End Time: 
Thursday, June 8, 2023 - 3:00pm
E2-506 or
Ricardo Sanfelice



Autonomy in robotics, transportation, and space applications requires resilient, fast, and safe motion planners. Specifically, we need fast motion planners that can safely navigate autonomous systems through cluttered environments, despite the limitations and uncertainty arising from low-cost onboard sensors and simplified mathematical models. Additionally, the generated motion plan must allow the autonomous system to abort its mission without compromising safety in the event of failure.

In this talk, I will describe some recent efforts to address these challenges using reachability and data. First, I will discuss a scalable multi-agent motion planner that combines reinforcement learning and constrained control theory to generate fast and safe motion plans under uncertainty. Second, I will present Safely, a single-agent motion planner for a robot with limited onboard sensing capabilities. Safely uses data, stochastic reachability, and sensitivity analysis to prescribe a safe motion plan under uncertainty and identifies trajectory-relevant obstacles for the sensing-constrained robot to sense at each time step. I will also present the results from hardware experiments for both of these works. If time permits, I will also discuss a stochastic reachability-based approach for abort-safe spacecraft rendezvous under actuation and navigational uncertainty.


Speaker Bio:

Abraham Vinod is a Research Scientist at Mitsubishi Electric Research Laboratories (MERL). He received his Bachelor’s and Master’s degrees in Electrical Engineering at the Indian Institute of Technology Madras (IITM) and his Ph.D. degree in Electrical Engineering from the University of New Mexico. Before joining MERL, he held a postdoctoral position at the University of Texas at Austin (UT Austin). His research broadly focuses on learning, planning, and decision-making under uncertainty for autonomous systems. His work won the Best Student Paper Award at the 2017 ACM Hybrid Systems: Computation and Control Conference, the Best Paper Award finalist at the 2018 ACM Hybrid Systems: Computation and Control Conference, and his Master's thesis won the Prof. Achim Bopp Prize (IITM).