---
title: "交互式奖励代理：通过环境状态验证实现GUI任务评估"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.25904v1"
permalink: "https://feed.astralor.com/items/itm_87590703bda1428667a01443"
published: "2026-07-28T16:01:38.000Z"
score: 79
channel: "ai"
tags: ["GUI-RewardBench","GUI代理","交互式奖励代理","任务评估","环境状态验证"]
---

# 交互式奖励代理：通过环境状态验证实现GUI任务评估

## AI 点评

> 这篇论文解决了GUI评估中环境状态获取的痛点，提出的IRA方法兼具创新性和实用性。对于从事GUI自动化、AI Agent训练的研究者和工程师来说，IRA不仅提供了一种可靠的奖励信号，也启发了如何将系统工具与界面信息结合进行验证。值得关注其在更复杂跨应用场景下的扩展能力。

## 速览

本文提出交互式奖励代理（IRA），基于“提出-验证”框架，在GUI代理执行后通过系统工具、应用工具和GUI工具交互获取并验证环境状态证据，以判断任务是否完成。同时构建了包含321条GUI任务轨迹的基准GUI-RewardBench。IRA在该基准上达到86.9%的准确率，并成功应用于强化学习训练，使OSWorld成功率提升至34.0%。

## 关键点

- 提出IRA框架，通过交互式环境状态验证解决GUI任务评估难题
- 构建GUI-RewardBench基准，覆盖10个Ubuntu应用类别共321条轨迹
- IRA准确率达86.9%，超越现有评估器基线
- IRA可作为奖励信号训练GUI代理，提升OSWorld成功率至34.0%

## 全文

Graphical user interface task evaluation aims to determine whether a GUI agent has successfully completed a user instruction. Automated GUI task evaluation has received increasing attention because the evaluation results can serve as reward signals for both test-time scaling and post-training. However, reliable GUI task evaluation remains challenging because the judgments often require access to environment states, such as system configurations, file data, and application settings, beyond the screenshots of execution trajectories. In this paper, we propose an interactive reward agent (IRA) based on a propose-then-verify framework to acquire and verify evidence from the post-execution environment. Given a task instruction and a GUI environment after the GUI agent execution, IRA first proposes the task completion conditions and then verifies them by invoking system tools, application tools, and GUI tools. This design combines evidence from both visible interfaces and the environment state in an interactive process. We further introduce GUI-RewardBench, a benchmark of 321 GUI task trajectories spanning 10 Ubuntu desktop application categories. Experiments show that IRA achieves 86.9% accuracy on GUI-RewardBench, outperforming existing evaluator baselines. We further apply IRA to reinforcement learning of GUI agents, achieving a 34.0% OSWorld success rate, which demonstrates that IRA can provide effective reward signals for training GUI agents.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_87590703bda1428667a01443