Dongbin Zhao

4 papers Institute of Automation, Chinese Academy of Sciences

• Equilibrium Policy Generalization: A Reinforcement Learning Framework for Cross-Graph Zero-Shot Generalization in Pursuit-Evasion Games
• Learning When to Think: Shaping Adaptive Reasoning in R1-Style Models via Multi-Stage RL
• Learning and Planning Multi-Agent Tasks via an MoE-based World Model
• Videos are Sample-Efficient Supervisions: Behavior Cloning from Videos via Latent Representations