- I am a Research Scientist at the Computer Science and Artificial Intelligence Laboratory (
CSAIL) at the Massachusetts Institute of Technology (MIT). I received my PhD in Control Systems/Electrical Engineering and PhD minor in Mathematics from the Pennsylvania State University. I also was a Postdoctoral Associate for two years with the Model-based Embedded and Robotic Systems (MERS) group at MIT's CSAIL. My goal is to develop new rigorous mathematical tools and algorithms to address challenging problems in Control Systems, Robotics, and Optimization. In particular, my research interests include probabilistic control, chance constrained optimization, stochastic systems, robotic systems, and machine learning. I lead a group of graduate students to develop risk bounded motion planning algorithms for autonomous systems such as self-driving vehicles, underwater vehicles, autonomous airplanes, and robotic arms. For related publications check here and for related tutorials check rarnop.mit.edu
- TOYOTA Project: GEORDI: Proactive and Risk-Bounded Autonomous Driving
GEORDI is an autonomous system that generates “proactive” and “risk-bounded” maneuvers for autonomous driving. GEORDI identifies potential risky traffic situations early on, well before they become imminent. GEORDI has the knowledge of risk levels for driving maneuvers where risk is defined as the probability of violating the safety constraints, e.g., the probability of collision. To be able to generate proactive and risk-aware maneuvers for the vehicle, we model the driving problem as a stochastic multi-agent game with one controllable player (ego vehicle) and multiple uncontrollable players (agent vehicles) each of which can choose from a set of maneuver actions. We, then, explore the belief space, e.g., space of probability distributions defined over the states of vehicles, to find a sequence of maneuvers for the ego vehicle to get from the initial state to the goal state. To ensure safety, we introduce chance constraints that bound the probability of failure of the obtained plan. To explore the belief space, we leverage our Risk-bounded AO* (RAO*) heuristic search algorithm, which is a probabilistic conditional planner. RAO* will be supported by 3 key components: i) Intention Recognition System to predict the sequence of maneuvers that other vehicles are likely to take. ii) Risk-Aware Motion Planer to design different driving maneuvers and controllers in the presence of probabilistic uncertainties considering the vehicle dynamics, and iii) Risk-Aware Maneuver Executive to estimate and monitor the probability of failure of maneuvers at execution time.