WebOperator: Action-Aware Tree Search for Autonomous Agents in Web Environment

2025-12-14

AI Summary

To address the lack of long-term planning and safe backtracking capabilities in LLM-based agents operating within partially observable web environments, this paper proposes an action-aware tree search framework. The framework introduces a novel robust backtracking mechanism grounded in feasibility verification, explicitly modeling action reversibility; integrates reward estimation with safety assessment to enable best-first search; and constructs high-quality action candidates via multi-context generation, DOM-aware filtering, and semantic normalization. Evaluated on the WebArena benchmark, the GPT-4o–based instantiation achieves a 54.6% task success rate—setting a new state-of-the-art. It significantly improves stability and error-correction robustness on complex navigation tasks, particularly excelling in scenarios involving irreversible actions.

Technology Category

• Search and Optimization: Learning to Search
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Multiagent Systems: Multiagent Planning

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

LLM-based agents often operate in a greedy, step-by-step manner, selecting actions solely based on the current observation without considering long-term consequences or alternative paths. This lack of foresight is particularly problematic in web environments, which are only partially observable-limited to browser-visible content (e.g., DOM and UI elements)-where a single misstep often requires complex and brittle navigation to undo. Without an explicit backtracking mechanism, agents struggle to correct errors or systematically explore alternative paths. Tree-search methods provide a principled framework for such structured exploration, but existing approaches lack mechanisms for safe backtracking, making them prone to unintended side effects. They also assume that all actions are reversible, ignoring the presence of irreversible actions-limitations that reduce their effectiveness in realistic web tasks. To address these challenges, we introduce WebOperator, a tree-search framework that enables reliable backtracking and strategic exploration. Our method incorporates a best-first search strategy that ranks actions by both reward estimates and safety considerations, along with a robust backtracking mechanism that verifies the feasibility of previously visited paths before replaying them, preventing unintended side effects. To further guide exploration, WebOperator generates action candidates from multiple, varied reasoning contexts to ensure diverse and robust exploration, and subsequently curates a high-quality action set by filtering out invalid actions pre-execution and merging semantically equivalent ones. Experimental results on WebArena and WebVoyager demonstrate the effectiveness of WebOperator. On WebArena, WebOperator achieves a state-of-the-art 54.6% success rate with gpt-4o, underscoring the critical advantage of integrating strategic foresight with safe execution.

Problem

• Addresses lack of foresight in web agents' step-by-step action selection
• Solves unsafe backtracking and irreversible action issues in tree-search methods
• Enables strategic exploration with safe execution in partially observable web environments

Innovation

• Tree-search framework with safe backtracking mechanism
• Best-first search ranking actions by reward and safety
• Generates diverse action candidates from multiple reasoning contexts