Session: 20-17-01: Rising Stars of Mechanical Engineering

Paper Number: 172155

Interactive Autonomy: Learning and Control for Multi-Agent Interactions 

My research focuses on building the algorithmic and mathematical foundations that enable safe, intelligent, and efficient interactions between robots and other agents in complex, shared environments. As robots become increasingly integrated into our daily lives, whether in the form of autonomous cars navigating roads with pedestrians and human drivers, delivery drones operating in urban airspaces, or mobile robots moving through warehouses, they must reason not just about the physical world, but also about the intentions, objectives, and behaviors of other agents in the environment. These application domains are fundamentally multi-agent in nature, and achieving reliable robotic behavior in such settings requires a principled approach to modeling and decision-making under interaction and uncertainty.

To address these challenges, my work brings together game-theoretic planning and control, risk-aware reasoning, and control-theoretic methods to design robotic systems that can operate robustly in real-world multi-agent environments. Game theory provides a powerful framework for capturing the interdependence of agents’ decisions: to act intelligently, a robot must anticipate how its own choices affect the decisions of others and vice versa. I develop models and algorithms that allow robots to reason about these dependencies and plan actions accordingly. A key focus of my work is identifying and exploiting structural properties in multi-agent interactions, such as local coupling or sparsity, to enable efficient planning algorithms that are suitable for real-time operation on robotic hardware.

Beyond planning, safety remains a central concern in deploying autonomous systems in the wild. To this end, I incorporate control-theoretic tools and formal safety guarantees into my algorithms, ensuring that robots not only behave intelligently but also avoid unsafe situations even in uncertain and dynamic environments. My research also explores risk-aware decision-making, where robots weigh the potential consequences of different outcomes and adapt their behavior accordingly, especially in safety-critical contexts.

Another major thrust of my work involves enabling robots to learn about the surrounding agents through observation. While classical inverse reinforcement learning can infer the preferences of isolated agents, real-world agents interact and adapt to one another. I develop mathematical models and numerical algorithms for inferring interdependent preferences from observed multi-agent behavior, allowing robots to better predict and respond to human or robot teammates and opponents.

Overall, my research aims to build a theoretical and algorithmic foundation that enables the deployment of autonomous systems capable of operating safely and effectively in the complex, interactive, and multi-agent settings that define many of today’s and tomorrow’s robotics applications.

Presenting Author: Negar Mehr UC Berkeley

Presenting Author Biography: Negar Mehr is an assistant professor in the Department of Mechanical Engineering at the University of California, Berkeley. She was an assistant professor of Aerospace Engineering at the University of Illinois Urbana-Champaign where she was also affiliated with the Coordinated Science Laboratory (CSL) and Electrical and Computer Engineering department at UIUC. She was a postdoctoral scholar at Stanford Aeronautics and Astronautics department from 2019 to 2020. She received her Ph.D. in Mechanical Engineering from UC Berkeley in 2019 and her B.Sc. in Mechanical Engineering from Sharif University of Technology, Tehran, Iran, in 2013. She is a recipient of the NSF CAREER Award. She was awarded the IEEE Intelligent Transportation Systems best Ph.D. dissertation award in 2020. Negar was recognized as a rising star in EECS, Aeronautics & Astronautics, and Civil and Environmental Engineering.

Authors: