Environment Maps: Structured Environmental Representations for Long-Horizon Agents

TL;DR

This paper introduces Environment Maps, a structured, persistent environmental representation that enhances long-horizon agent planning by consolidating diverse evidence into a graph, improving success rates in complex tasks.

Contribution

The paper presents Environment Maps, a novel structured environmental representation that improves long-horizon agent performance by integrating heterogeneous evidence into a persistent, editable graph.

Findings

Agents with Environment Maps achieve 28.2% success rate, nearly doubling baseline performance.

Environment Maps outperform raw trajectory data access, which yields 23.3% success.

Structured representations enable more robust, interpretable, and incremental long-horizon planning.

Abstract

Although large language models (LLMs) have advanced rapidly, robust automation of complex software workflows remains an open problem. In long-horizon settings, agents frequently suffer from cascading errors and environmental stochasticity; a single misstep in a dynamic interface can lead to task failure, resulting in hallucinations or trial-and-error. This paper introduces $\textit{Environment Maps}$: a persistent, agent-agnostic representation that mitigates these failures by consolidating heterogeneous evidence, such as screen recordings and execution traces, into a structured graph. The representation consists of four core components: (1) Contexts (abstracted locations), (2) Actions (parameterized affordances), (3) Workflows (observed trajectories), and (4) Tacit Knowledge (domain definitions and reusable procedures). We evaluate this framework on the WebArena benchmark across five domains. Agents equipped with environment maps achieve a 28.2% success rate, nearly doubling the performance of baselines limited to session-bound context (14.2%) and outperforming agents that have access to the raw trajectory data used to generate the environment maps (23.3%). By providing a structured interface between the model and the environment, Environment Maps establish a persistent foundation for long-horizon planning that is human-interpretable, editable, and incrementally refinable.