Reflection-Based Memory For Web navigation Agents
🤖 AI Summary
Existing web navigation agents lack long-term memory mechanisms, hindering the reuse of historical experience and leading to repeated failures. To address this, we propose Reflection-Augment Planning (ReAP), a novel framework that introduces a reflective memory mechanism—integrating LLM-based self-reflection (to generate causal attributions for task success or failure), structured experience storage, and retrieval-augmented planning. This enables cross-task knowledge transfer and adaptive decision-making. Empirical evaluation demonstrates that ReAP improves overall navigation performance by 11 points and achieves a substantial 29-point gain on previously failed tasks, validating its effectiveness in enhancing generalization and error correction capabilities.
Application Category
📝 Abstract
Web navigation agents have made significant progress, yet current systems operate with no memory of past experiences -- leading to repeated mistakes and an inability to learn from previous interactions. We introduce Reflection-Augment Planning (ReAP), a web navigation system to leverage both successful and failed past experiences using self-reflections. Our method improves baseline results by 11 points overall and 29 points on previously failed tasks. These findings demonstrate that reflections can transfer to different web navigation tasks.
Problem
Research questions and friction points this paper is trying to address.
Web agents lack memory for past experiences
Repeated mistakes hinder learning from interactions
Reflection-Augment Planning improves task performance
Innovation
Methods, ideas, or system contributions that make the work stand out.
Reflection-Augment Planning (ReAP) system
Leverages past successful and failed experiences
Improves baseline results by 11-29 points
🔎 Similar Papers
No similar papers found.
