Mirage-1: Augmenting and Updating GUI Agent with Hierarchical Multimodal Skills

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Xie, Yuquan, Li, Zaijing, Shao, Rui, Chen, Gongwei, Zhou, Kaiwen, Li, Yinchuan, Jiang, Dongmei, Nie, Liqiang
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866911001843073024
author Xie, Yuquan
Li, Zaijing
Shao, Rui
Chen, Gongwei
Zhou, Kaiwen
Li, Yinchuan
Jiang, Dongmei
Nie, Liqiang
author_facet Xie, Yuquan
Li, Zaijing
Shao, Rui
Chen, Gongwei
Zhou, Kaiwen
Li, Yinchuan
Jiang, Dongmei
Nie, Liqiang
contents Recent efforts to leverage the Multi-modal Large Language Model (MLLM) as GUI agents have yielded promising outcomes. However, these agents still struggle with long-horizon tasks in online environments, primarily due to insufficient knowledge and the inherent gap between offline and online domains. In this paper, inspired by how humans generalize knowledge in open-ended environments, we propose a Hierarchical Multimodal Skills (HMS) module to tackle the issue of insufficient knowledge. It progressively abstracts trajectories into execution skills, core skills, and ultimately meta-skills, providing a hierarchical knowledge structure for long-horizon task planning. To bridge the domain gap, we propose the Skill-Augmented Monte Carlo Tree Search (SA-MCTS) algorithm, which efficiently leverages skills acquired in offline environments to reduce the action search space during online tree exploration. Building on HMS, we propose Mirage-1, a multimodal, cross-platform, plug-and-play GUI agent. To validate the performance of Mirage-1 in real-world long-horizon scenarios, we constructed a new benchmark, AndroidLH. Experimental results show that Mirage-1 outperforms previous agents by 32\%, 19\%, 15\%, and 79\% on AndroidWorld, MobileMiniWob++, Mind2Web-Live, and AndroidLH, respectively. Project page: https://cybertronagent.github.io/Mirage-1.github.io/
format Preprint
id arxiv_https___arxiv_org_abs_2506_10387
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Mirage-1: Augmenting and Updating GUI Agent with Hierarchical Multimodal Skills
Xie, Yuquan
Li, Zaijing
Shao, Rui
Chen, Gongwei
Zhou, Kaiwen
Li, Yinchuan
Jiang, Dongmei
Nie, Liqiang
Artificial Intelligence
Recent efforts to leverage the Multi-modal Large Language Model (MLLM) as GUI agents have yielded promising outcomes. However, these agents still struggle with long-horizon tasks in online environments, primarily due to insufficient knowledge and the inherent gap between offline and online domains. In this paper, inspired by how humans generalize knowledge in open-ended environments, we propose a Hierarchical Multimodal Skills (HMS) module to tackle the issue of insufficient knowledge. It progressively abstracts trajectories into execution skills, core skills, and ultimately meta-skills, providing a hierarchical knowledge structure for long-horizon task planning. To bridge the domain gap, we propose the Skill-Augmented Monte Carlo Tree Search (SA-MCTS) algorithm, which efficiently leverages skills acquired in offline environments to reduce the action search space during online tree exploration. Building on HMS, we propose Mirage-1, a multimodal, cross-platform, plug-and-play GUI agent. To validate the performance of Mirage-1 in real-world long-horizon scenarios, we constructed a new benchmark, AndroidLH. Experimental results show that Mirage-1 outperforms previous agents by 32\%, 19\%, 15\%, and 79\% on AndroidWorld, MobileMiniWob++, Mind2Web-Live, and AndroidLH, respectively. Project page: https://cybertronagent.github.io/Mirage-1.github.io/
title Mirage-1: Augmenting and Updating GUI Agent with Hierarchical Multimodal Skills
topic Artificial Intelligence
url https://arxiv.org/abs/2506.10387