I am an Affiliated Faculty at the British Open-ended Learning and Discovery (BOLD) Lab, a Visiting Researcher in the Department of Engineering Science at the University of Oxford, hosted by Prof. Jakob Foerster, and a Research Professor in Transportation Engineering at Tongji University. Previously, I was a Postdoctoral Fellow at the Harvard Kennedy School. I received a Ph.D. in Transportation Engineering from New York University, an M.Sc. from Tsinghua University, and a B.Sc. from Jilin University.
Research
My research sits at the intersection of control, optimization, and machine learning, building toward embodied and physical AI. I develop principled decision-making frameworks—spanning reinforcement learning, optimal control, robot foundation models, and embodied AI agents—that bridge rigorous theoretical guarantees with high-performance adaptive intelligence. By addressing the fundamental trade-offs between efficiency, safety, and generalizability, my work seeks to empower autonomous systems—robots, autonomous vehicles, and large-scale mobility networks—to perceive, reason, and act reliably in complex, uncertain, and dynamic environments.
Reinforcement Learning
The theoretical and algorithmic foundations of offline RL and multi-agent RL (MARL), including zero-shot coordination.
- Focus: Reliable policy learning from static datasets, and coordinating multiple agents—including zero-shot coordination with unseen partners.
- Approach: Uncertainty quantification with game-theoretic and population-based methods for safe, robust, and generalizable policies.
Control & Optimization
The analytical foundations of optimal control, dynamic programming, and stochastic optimization for multi-agent and networked systems.
- Focus: Structural properties, stability, and coordination across temporal and spatial scales.
- Approach: Translating this analysis into intelligent transportation networks, distributed control, and autonomous-fleet optimization.
Robot Foundation Models
Foundation models for robotics—multi-modal world models and vision-language-action (VLA) policies—for scalable decision-making.
- Focus: Multi-modal world models that learn environment dynamics for prediction and planning, and VLA models that map perception and language to action.
- Approach: Large-scale multimodal pretraining combined with model-based planning to generalize across tasks and embodiments.
Embodied AI Agents
Embodied agents that perceive, reason, and act in the physical world, closing the loop between learning and control.
- Focus: Long-horizon decision-making and coordination for physical agents under real-world constraints.
- Approach: Uniting foundation models, planning, and control for robust, data-efficient embodied intelligence.
People
Current Graduate Students
Former Members
Selected Publications
- Xiong, X., Liu, L., Li, Y., and Foerster, J.N. Bayesian Offline Multi-Agent Reinforcement Learning with Latent World Models. Preprint, 2026. arXiv
- Li, J., Cao, L., Liu, Q., and Xiong, X. Dreamer-SAC: Off-Policy Learning in Latent World Models for Sample-Efficient Autonomous Driving. In Transportation Research Board 106th Annual Meeting, TRB 2027. arXiv
- Pang, A., Wang, M., Pun, M., Chen, C., and Xiong, X. iLLM-TSC: Integration Reinforcement Learning and Large Language Model for Traffic Signal Control Policy Improvement. IEEE Transactions on Vehicular Technology, 2026. link
- Bai, Y., Li, Y., and Xiong, X. Multi-armed Bandit for Stochastic Shortest Path in Mixed Autonomy. In 28th IEEE International Conference on Intelligent Transportation Systems, ITSC 2025. link
- Liu, L., Xie, C., and Xiong, X. Optimizing Highway Traffic Flow in Mixed Autonomy: A Multiagent Truncated Rollout Approach. Transportation Research Part B: Methodological (under review). arXiv
- Xiong, X., Wang, M., Sun, D., and Jin, L. An Approximate Dynamic Programming Approach to Vehicle Platooning Coordination in Networks. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(11): 16536–16547. link
- Xiong, X., Sha, J. and Jin, L. Optimizing Coordinated Vehicle Platooning: An Analytical Approach Based on Stochastic Dynamic Programming. Transportation Research Part B: Methodological, 2021, 150: 482–502. link
See the full publication list or Google Scholar.