![[2502.03752] Self-Improving Skill Learning for Robust Skill-based Meta-Reinforcement Learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2502.03752] Self-Improving Skill Learning for Robust Skill-based Meta-Reinforcement Learning
Summary
This paper presents Self-Improving Skill Learning (SISL), a novel approach to enhance skill-based meta-reinforcement learning by refining skills through self-guided policies, addressing challenges posed by noisy data.
Why It Matters
The research addresses critical issues in meta-reinforcement learning, particularly the instability caused by noisy offline demonstrations. By proposing SISL, the authors provide a solution that enhances the robustness and adaptability of skill learning in complex environments, which is essential for advancing AI applications in real-world scenarios.
Key Takeaways
- SISL improves skill-based meta-reinforcement learning by refining skills through self-guided policies.
- The approach mitigates the impact of noise in offline demonstrations, leading to more stable learning.
- SISL prioritizes task-relevant trajectories for skill updates, enhancing performance in long-horizon tasks.
- The method consistently outperforms existing skill-based meta-RL techniques.
- Code for SISL is publicly available, promoting further research and application.
Computer Science > Machine Learning arXiv:2502.03752 (cs) [Submitted on 6 Feb 2025 (v1), last revised 19 Feb 2026 (this version, v4)] Title:Self-Improving Skill Learning for Robust Skill-based Meta-Reinforcement Learning Authors:Sanghyeon Lee, Sangjun Bae, Yisak Park, Seungyul Han View a PDF of the paper titled Self-Improving Skill Learning for Robust Skill-based Meta-Reinforcement Learning, by Sanghyeon Lee and 3 other authors View PDF HTML (experimental) Abstract:Meta-reinforcement learning (Meta-RL) facilitates rapid adaptation to unseen tasks but faces challenges in long-horizon environments. Skill-based approaches tackle this by decomposing state-action sequences into reusable skills and employing hierarchical decision-making. However, these methods are highly susceptible to noisy offline demonstrations, leading to unstable skill learning and degraded performance. To address this, we propose Self-Improving Skill Learning (SISL), which performs self-guided skill refinement using decoupled high-level and skill improvement policies, while applying skill prioritization via maximum return relabeling to focus updates on task-relevant trajectories, resulting in robust and stable adaptation even under noisy and suboptimal data. By mitigating the effect of noise, SISL achieves reliable skill learning and consistently outperforms other skill-based meta-RL methods on diverse long-horizon tasks. Our code is available at this https URL. Comments: Subjects: Machine Learning (cs.LG)...
