AgentOPSD: Recursive Self-Distillation for Agentic Reinforcement Learning

2026-08-06

AI Summary

In long-horizon, multi-turn agent tasks, sparse terminal rewards pose significant challenges for attributing credit to critical decision steps. This work proposes a critic-free recursive self-distillation method that constructs episode-level evidence from token-level teacher-student log-probability differences and recursively updates a Bayesian belief state in log-odds space, enabling episode-level credit assignment without auxiliary critics or additional sampling. By innovatively integrating recursive Bayesian belief updating with self-distillation, the approach generates high-quality credit signals directly from sparse outcomes. Experiments demonstrate substantial improvements over GRPO and existing self-distillation baselines on ALFWorld, WebShop, and Search-QA, with Qwen2.5-7B achieving a success rate of 89.1% on ALFWorld.

Abstract

Reinforcement learning (RL) with verifiable rewards constructs trajectory-level advantage estimates, yet it often fails to credit the few pivotal decisions that determine outcomes in long-horizon, multi-turn agentic tasks. Recent work introduces privileged self-distillation for credit assignment, providing denser supervision, but it remains unclear how such local signals should represent sequential credit. We propose AgentOPSD, a critic-free, recursive method for turn-level credit assignment in agentic reinforcement learning. AgentOPSD aggregates token-level teacher-student log-probability gaps into turn-level evidence and recursively updates a Bayesian belief state in log-odds space. This yields a principled reweighting scheme that converts sparse outcome supervision into turn-level credit signals and identifies pivotal turns through the marginal belief revision between consecutive states. The method is fully compatible with standard policy optimization and requires neither an additional critic nor extra rollouts. We evaluate AgentOPSD on ALFWorld, WebShop, and Search-QA using Qwen2.5 models at two scales (3B and 7B). AgentOPSD outperforms GRPO and strong self-distillation baselines, achieving 89.1% success on ALFWorld with Qwen2.5-7B. Ablation studies attribute the gains to turn-level aggregation and history-dependent recursive belief updates.

Problem

• credit assignment
• agentic reinforcement learning
• long-horizon tasks
• turn-level credit
• sparse rewards

Innovation

• recursive self-distillation
• turn-level credit assignment
• Bayesian belief update
• critic-free reinforcement learning
• agentic RL