Jingqi Li

Advancing Safe, Strategic Autonomy under Information Asymmetry.

I'm a Peter O' Donnell Jr. Postdoctoral Fellow at the Oden Institute for Computational Engineering & Sciences of UT Austin, where I'm advised by Professor David Fridovich-Keil. I completed my Ph.D. in EECS from UC Berkeley in 2025, co-advised by Professor Claire Tomlin and Professor Somayeh Sojoudi. I was a member of Berkeley AI Research (BAIR) Lab. 

About my research

My work centers on dynamic game theory, control, and reinforcement learning. I develop theory and algorithms that enhance agents’ strategic planning, enabling them to learn and coordinate in complex environments. Applications include autonomous driving, air mobility, multi-robot manipulation, and smart grids. Recently, I focus on:

1. Developing efficient computation methods -- including model-based mathematical programming, data-driven model-free learning, and hybrid approaches -- that enable safe, decentralized, game-theoretic decision-making;

2. Characterizing and exploiting information asymmetry to supercharge decentralized autonomy:  We aim to address a crucial bottleneck that agents face when going from a single-agent mindset to a multi-agent setting: no single agent can dictate to everyone. Thus, agents must anticipate others and learn to coordinate effectively despite persistent information asymmetry. 

Selected works:

• [SIOPT '24] We proposed a game-theoretic primal-dual interior-point method for multiplayer nonlinear feedback Stackelberg games. Our method is provably convergent, and it can handle conflicting objectives and  coupled constraints among strategic players. 

• [RA-L '25] We developed a reachability learning framework offering deterministic safety guarantees for nonlinear systems under uncertainty, extending beyond the probabilistic safety guarantees of prior works. Our method was verified in drone racing hardware experiments, where it outperformed human pilots.

• [AAMAS '23] We introduced an efficient gradient approximator for the likelihood function of inverse nonlinear games, enabling fast gradient-based intent inference.

(I'm happy to collaborate on other exciting problems in game-theoretic control and learning, and so please feel free to reach out!)