![[2602.13949] Experiential Reinforcement Learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.13949] Experiential Reinforcement Learning
Summary
The paper introduces Experiential Reinforcement Learning (ERL), a new paradigm that enhances learning efficiency in language models by integrating self-reflection into the reinforcement learning process.
Why It Matters
ERL addresses the challenges of sparse and delayed feedback in reinforcement learning, providing a structured approach to improve learning outcomes. This innovation could significantly enhance the performance of AI systems in complex environments, making it a crucial development in the field of machine learning.
Key Takeaways
- Experiential Reinforcement Learning (ERL) embeds a reflection loop in the learning process.
- ERL improves exploration and stabilizes optimization in language models.
- The approach leads to significant performance gains, up to +81% in complex environments.
- Self-reflection in policy training transforms feedback into durable behavioral improvements.
- ERL demonstrates enhanced learning efficiency over traditional reinforcement learning methods.
Computer Science > Machine Learning arXiv:2602.13949 (cs) [Submitted on 15 Feb 2026] Title:Experiential Reinforcement Learning Authors:Taiwei Shi, Sihao Chen, Bowen Jiang, Linxin Song, Longqi Yang, Jieyu Zhao View a PDF of the paper titled Experiential Reinforcement Learning, by Taiwei Shi and 5 other authors View PDF HTML (experimental) Abstract:Reinforcement learning has become the central approach for language models (LMs) to learn from environmental reward or feedback. In practice, the environmental feedback is usually sparse and delayed. Learning from such signals is challenging, as LMs must implicitly infer how observed failures should translate into behavioral changes for future iterations. We introduce Experiential Reinforcement Learning (ERL), a training paradigm that embeds an explicit experience-reflection-consolidation loop into the reinforcement learning process. Given a task, the model generates an initial attempt, receives environmental feedback, and produces a reflection that guides a refined second attempt, whose success is reinforced and internalized into the base policy. This process converts feedback into structured behavioral revision, improving exploration and stabilizing optimization while preserving gains at deployment without additional inference cost. Across sparse-reward control environments and agentic reasoning benchmarks, ERL consistently improves learning efficiency and final performance over strong reinforcement learning baselines, achieving...
