SpiritSight Agent: Advanced GUI Agent with One Look

2025-03-05

AI Summary

Current vision-based GUI agents suffer from significant bottlenecks in element localization accuracy, hindering real-world deployment. This paper proposes an end-to-end vision-driven GUI agent targeting imprecise element localization and input ambiguity under high-resolution cross-platform interfaces. Our approach introduces three key contributions: (1) GUI-Lasagne—the first hierarchical, structured GUI dataset supporting fine-grained spatial reasoning; (2) Universal Block Parsing (UBP), a novel algorithm enabling pixel-accurate element grounding and dynamic resolution-robust modeling; and (3) joint optimization of state-of-the-art vision-language models with large-scale synthetic GUI data. Evaluated across multi-platform benchmarks, our method achieves a 23.6% average improvement in localization accuracy while maintaining low latency (<380 ms) and strong cross-platform compatibility. To our knowledge, this is the first work to unify high-precision localization, low-latency inference, and robust cross-platform navigation in a single GUI agent framework.

Technology Category

• Humans and AI: Intelligent User Interfaces
• Computer Vision: Biometrics, Face, Gesture & Pose
• Intelligent Robots: Localization, Mapping, and Navigation

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Agentic search
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Graphical User Interface (GUI) agents show amazing abilities in assisting human-computer interaction, automating human user's navigation on digital devices. An ideal GUI agent is expected to achieve high accuracy, low latency, and compatibility for different GUI platforms. Recent vision-based approaches have shown promise by leveraging advanced Vision Language Models (VLMs). While they generally meet the requirements of compatibility and low latency, these vision-based GUI agents tend to have low accuracy due to their limitations in element grounding. To address this issue, we propose $ extbf{SpiritSight}$, a vision-based, end-to-end GUI agent that excels in GUI navigation tasks across various GUI platforms. First, we create a multi-level, large-scale, high-quality GUI dataset called $ extbf{GUI-Lasagne}$ using scalable methods, empowering SpiritSight with robust GUI understanding and grounding capabilities. Second, we introduce the $ extbf{Universal Block Parsing (UBP)}$ method to resolve the ambiguity problem in dynamic high-resolution of visual inputs, further enhancing SpiritSight's ability to ground GUI objects. Through these efforts, SpiritSight agent outperforms other advanced methods on diverse GUI benchmarks, demonstrating its superior capability and compatibility in GUI navigation tasks. Models are available at $href{https://huggingface.co/SenseLLM/SpiritSight-Agent-8B}{this URL}$.

Problem

• Improves accuracy of vision-based GUI agents
• Enhances GUI object grounding in dynamic environments
• Ensures compatibility across diverse GUI platforms

Innovation

• Vision-based end-to-end GUI agent
• Multi-level large-scale GUI dataset
• Universal Block Parsing for ambiguity resolution