1. Equilibrium Finding in Multi-Agent Systems

  • Description: Core research focusing on theoretically rigorous and practically applicable algorithms for finding Nash equilibria in complex stochastic games. This work led to the development of methods like Constrained Exploitability Descent (CED) and Divergence-Regularized Discounted Aggregation (DRDA).
  • Key Outcomes: Converging algorithms designed specifically for integration with Deep RL and large-scale function approximation.
  • Related Publications: ICML 2025, ICLR 2025, IEEE TNNLS 2024.

2. Equilibrium Policy Generalization (EPG)

  • Description: Developed a novel RL paradigm for deriving real-time robust strategies in adversarial games (e.g., pursuit-evasion games) under dynamically changing environments (e.g., varying graph structures).
  • Key Outcomes: Achieved desirable cross-graph zero-shot generalization performance, critical for real-world robotics and security applications.
  • Related Publications: NeurIPS 2025.

3. Hierarchical Campus Security System

  • Description: An applied project focused on enabling real-time, robust decision-making for security robots. The system includes modules for sensor allocation and worst-case robust robotic pursuit under partial observability.
  • Key Outcomes: Development of the R2PS approach, allowing for real-time strategy generation in dynamical security scenarios.
  • Related Publications: ICLR 2026 (and an upcoming manuscript).

Teaching and Mentoring

  • Teaching Assistant (2023 – 2026): Reinforcement Learning (Graduate Course), UCAS.
  • Teaching Assistant (2021): Design and Analysis of Algorithms Lab (Undergraduate Course), UCAS.