Research Interests

• Intelligent Systems

• Reinforcement Learning

• Generative AI

• Differential Equations

• Optimal Control

• Network Optimization

Scalable and Robust Decision-Making via World Foundation Models and Reinforcement Learning

This project aims to develop an intelligent autonomous driving system by integrating reinforcement learning (RL) with world foundation models. RL will provide adaptive decision-making in complex environments, while the world models—trained on diverse, multimodal data—will simulate realistic driving scenarios to enhance planning and generalization. These world models will capture spatial-temporal dynamics of road users, infrastructure, and environmental contexts, allowing autonomous agents to anticipate and respond to diverse real-world conditions. The integration enables efficient policy learning with reduced dependence on costly real-world data, promoting safety and robustness. The ultimate goal is to create a scalable and generalizable framework for autonomous vehicles that can learn, plan, and act safely in dynamic and uncertain driving environments.

Sponsor: Fundamental Research Funds for the Central Universities

Statistical and Optimization Study of Efficient Machine Learning for High-Dimensional and Streaming Data

This research aims to advance machine learning by developing novel optimization techniques and statistical tools for effectively handling high-dimensional and streaming data. We propose an integrated framework combining statistics, optimization, and machine learning, incorporating advanced methods such as Folded Concave Penalty (FCP) regularization, non-asymptotic analysis, higher-order optimization, distributed algorithms, and online learning for high-dimensional multi-armed bandits and reinforcement learning. Our work will improve computational efficiency in non-convex optimization and achieve optimal statistical error bounds. By moving beyond L1 regularization and first-order solutions, this research opens new directions in high-dimensional statistical learning and online decision-making. Additionally, we will collaborate with the healthcare industry and AI companies to apply and validate these methods in practical, real-world scenarios.

Sponsor: National Natural Science Foundation of China
Collaborator: Prof. Tao Yao

Large-Scale Vehicle Routing with Reinforcement Learning and Large Language Models

This project aims to develop a large-scale vehicle routing system by integrating reinforcement learning (RL) with large language models (LLMs). By leveraging the decision-making capabilities of RL and the data processing power of LLMs, the system will dynamically optimize vehicle routes in real-time, adapting to complex and ever-changing traffic conditions. LLMs will interpret traffic data, analyze road conditions, and predict congestion patterns, while RL will continuously refine routing strategies based on real-world feedback. The ultimate goal is to create a scalable, intelligent routing system that enhances traffic efficiency, reduces congestion, and improves the system's ability to respond to unexpected events, offering a more adaptive and efficient solution for urban transportation networks.

Sponsor: Fundamental Research Funds for the Central Universities

Reinforcement Learning from Human Feedback for Lane Changing of Autonomous Vehicles in Mixed-Traffic

This project aims to address the lane-changing behavior of autonomous vehicles in mixed traffic environments to enhance road safety and traffic efficiency. By thoroughly analyzing the behavioral differences between human drivers and autonomous vehicles, we will design a human-like lane-changing decision system tailored to the characteristics of autonomous vehicles. This system will utilize reinforcement learning combined with human driver feedback, enabling autonomous vehicles to better understand and adapt to human driving preferences, thereby making safer and more efficient lane-changing decisions in complex traffic environments. We plan to validate the effectiveness and reliability of this system through simulations and field tests, further improving the predictability of autonomous vehicle behavior, interaction safety, and traffic efficiency. The outcomes of this research will provide crucial technical support for the future realization of safe and efficient autonomous driving systems, promote the widespread application of autonomous driving technology in road traffic, and offer new insights and solutions for solving real-world traffic problems.

Sponsor: Fundamental Research Funds for the Central Universities

Vehicle Platooning Coordination Across Multiple Junctions on Highway Networks

Connected and autonomous vehicles (CAVs) can enhance road traffic efficiency and reduce energy consumption. Current research, both domestically and internationally, mainly focuses on the microscopic vehicle level, lacking a macroscopic perspective on highway networks. Existing optimizations consider only intelligent connected vehicles’ decisions, ignoring interactions with manually driven vehicles and the impact of traditional traffic flows on control units at junction hubs. This project addresses these gaps by studying collaborative optimization of vehicle platooning across multiple junctions in cascading road segments. It describes the interaction mechanisms between intelligent and manually driven vehicles, employs data-driven methods for multi-junction decision-making, and integrates these with the physical model of optimal strategies at junction hubs to improve adaptability. The research, validated on a simulation platform with real road network data, offers new approaches and a theoretical foundation for optimizing vehicle platooning in complex highway networks.

Sponsor: Fundamental Research Funds for the Central Universities