Experience is the Best Teacher: Motivating Effective Exploration in Reinforcement Learning for LLMs

2026-03-20

AI Summary

This work addresses the challenge that large language models often struggle to discover high-reward behaviors in rule-based reinforcement learning due to exploration being confined within the current policy distribution. To overcome this limitation, the paper proposes HeRL, a novel framework that explicitly leverages hindsight experiences—specifically, suboptimal trajectories along with their evaluation criteria—as contextual guidance to steer policy exploration toward out-of-distribution, high-quality responses. HeRL integrates hindsight experience replay, in-context learning, and gradient-based expectation estimation, and incorporates reward shaping to prioritize outputs with greater potential for improvement. Evaluated across multiple benchmarks, HeRL significantly outperforms existing methods, enhancing both exploration efficiency and reasoning capability, while also demonstrating a test-time self-improvement property through experience-guided adaptation.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Reinforcement Learning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Reinforcement Learning (RL) with rubric-based rewards has recently shown remarkable progress in enhancing general reasoning capabilities of Large Language Models (LLMs), yet still suffers from ineffective exploration confined to curent policy distribution. In fact, RL optimization can be viewed as steering the policy toward an ideal distribution that maximizes the rewards, while effective exploration should align efforts with desired target. Leveraging this insight, we propose HeRL, a Hindsight experience guided Reinforcement Learning framework to bootstrap effective exploration by explicitly telling LLMs the desired behaviors specified in rewards. Concretely, HeRL treats failed trajectories along with their unmet rubrics as hindsight experience, which serves as in-context guidance for the policy to explore desired responses beyond its current distribution. Additionally, we introduce a bonus reward to incentivize responses with greater potential for improvement under such guidance. HeRL facilitates effective learning from desired high quality samples without repeated trial-and-error from scratch, yielding a more accurate estimation of the expected gradient theoretically. Extensive experiments across various benchmarks demonstrate that HeRL achieves superior performance gains over baselines, and can further benefit from experience guided self-improvement at test time. Our code is available at https://github.com/sikelifei/HeRL.

Problem

• Reinforcement Learning
• Large Language Models
• Exploration
• Reward Shaping
• Hindsight Experience

Innovation

• Hindsight Experience
• Reinforcement Learning
• Large Language Models
• Reward-Guided Exploration
• In-Context Learning