When Privileged Guidance Misaligns: State-Matched Routing and Contextualized Self-Distillation for Multi-Turn Agents
The paper introduces State-Matched Routing and Contextualized Self-Distillation (SMRC-SD) to improve multi-turn agents using privileged reference guidance. It solves state–reference mismatch by matching the agent's current state against successful trajectory states, ensuring the teacher only provides relevant guidance when a trajectory is applicable. This method outperforms unconditional distillation in various embodied and web tasks.