M$^2$: Dual-Memory Augmentation for Long-Horizon Web Agents via Trajectory Summarization and Insight Retrieval
It addresses a critical bottleneck in autonomous web agents for handling complex, long-horizon tasks, representing a novel method rather than incremental.
The paper tackles the problem of long-horizon web navigation by proposing M^2, a training-free, memory-augmented framework that improves success rates by up to 19.6% and reduces tokens by 58.7% for Qwen3-VL-32B.
Multimodal Large Language Models (MLLMs) based agents have demonstrated remarkable potential in autonomous web navigation. However, handling long-horizon tasks remains a critical bottleneck. Prevailing strategies often rely heavily on extensive data collection and model training, yet still struggle with high computational costs and insufficient reasoning capabilities when facing complex, long-horizon scenarios. To address this, we propose M$^2$, a training-free, memory-augmented framework designed to optimize context efficiency and decision-making robustness. Our approach incorporates a dual-tier memory mechanism that synergizes Dynamic Trajectory Summarization (Internal Memory) to compress verbose interaction history into concise state updates, and Insight Retrieval Augmentation (External Memory) to guide the agent with actionable guidelines retrieved from an offline insight bank. Extensive evaluations across WebVoyager and OnlineMind2Web demonstrate that M$^2$ consistently surpasses baselines, yielding up to a 19.6% success rate increase and 58.7% token reduction for Qwen3-VL-32B, while proprietary models like Claude achieve accuracy gains up to 12.5% alongside significantly lower computational overhead.