Retrieval-Augmented Skill Optimization via Cross-Harness Adaptation

{'Problem': 'Existing skill optimization methods overlook accumulated knowledge from publicly shared skills and rely on expensive agent rollouts. This paper addresses the gap in leveraging external knowledge to improve skill optimization efficiency, proposing a novel framework that integrates retrieval mechanisms to enhance skill learning.', 'Method': 'The authors propose Retrieval-Augmented Skill Optimization (RASO), which consists of two main components: Retrieval-Augmented Skill Initialization (RASI) and Retrieval-Augmented Skill Update (RASU). RASI constructs a knowledge-grounded initial skill by retrieving relevant information from external sources, eliminating the need for costly agent rollouts. RASU iteratively refines the skill by retrieving additional external knowledge, guided by execution feedback, thus allowing for continuous improvement of the skill based on real-time performance data.', 'Results': 'RASO consistently outperforms baselines that do not utilize retrieval-augmented skill initialization and updating. However, the specific baselines and quantitative performance metrics are not disclosed in the available text.', 'Limitations': 'The authors do not report any limitations in their work. However, the lack of specified baselines and quantitative results may hinder the ability to fully assess the performance improvements offered by RASO.', 'Why it matters': 'This work has significant implications for downstream applications in reinforcement learning and skill acquisition, as it demonstrates a method to efficiently leverage external knowledge, potentially reducing the computational burden associated with traditional skill optimization methods.'}