LASER: LLM Agent with State-Space Exploration for Web Navigation

TL;DR

This paper introduces LASER, a novel LLM-based web navigation method that models tasks as state space exploration, enabling error recovery and significantly improving performance over previous approaches.

Contribution

LASER reformulates web navigation as state space exploration, allowing flexible backtracking and error correction, which enhances LLM decision-making in complex scenarios.

Findings

LASER outperforms previous methods on web navigation tasks.

LASER closes the gap with human performance.

Experimental results demonstrate improved robustness and accuracy.

Abstract

Large language models (LLMs) have been successfully adapted for interactive decision-making tasks like web navigation. While achieving decent performance, previous methods implicitly assume a forward-only execution mode for the model, where they only provide oracle trajectories as in-context examples to guide the model on how to reason in the environment. Consequently, the model could not handle more challenging scenarios not covered in the in-context examples, e.g., mistakes, leading to sub-optimal performance. To address this issue, we propose to model the interactive task as state space exploration, where the LLM agent transitions among a pre-defined set of states by performing actions to complete the task. This formulation enables flexible backtracking, allowing the model to recover from errors easily. We evaluate our proposed LLM Agent with State-Space ExploRation (LASER) on both the WebShop task and amazon.com. Experimental results show that LASER significantly outperforms previous methods and closes the gap with human performance on the web navigation task.