Teach-and-Repeat: Accurately Extracting Operational Knowledge from Mobile Screen Demonstrations to Empower GUI Agents

2026-06-10

AI Summary

Current vision-language models struggle to accurately extract reusable operational semantics from heterogeneous mobile application interfaces, hindering the translation of screen demonstrations into automated execution. This work proposes a Teach-and-Repeat paradigm, wherein a Teach VLM automatically identifies keyframes from demonstration videos and generates structured natural language instructions to guide downstream GUI agents in task completion. The core contributions include the first method for extracting interpretable and reusable procedural knowledge directly from raw screen interaction traces, along with the construction of a Chinese mobile teaching benchmark and a data flywheel mechanism. Experiments demonstrate that Teach VLM achieves state-of-the-art performance in operational semantic prediction, significantly outperforming existing VLM baselines and consistently improving task success rates of agents in Android World.

Abstract

Understanding the digital world on mobile devices is shifting from static UI perception to dynamic action comprehension. This capability enables models to convert visual state transitions into operational knowledge, defined as short natural-language sentences that describe action types, target UI elements, textual arguments, and execution orders. However, due to the highly diverse and heterogeneous UI designs across applications, existing vision-language models (VLMs) struggle to accurately infer these underlying operations. To bridge this gap, we introduce Teach VLM, a core model designed to translate mobile screen trajectories into step-wise operational knowledge by extracting and analyzing operation-related keyframes from demonstration videos. To address the scarcity of aligned training data, we develop a systematic data flywheel for scalable data acquisition. We further introduce a novel Chinese Mobile Screen Teach Benchmark for fine-grained evaluation. Building upon Teach VLM, we propose the Teach-and-Repeat paradigm, where the generated operational knowledge serves as an interpretable procedural reference to guide downstream screen-based execution agents. Extensive evaluations demonstrate that Teach VLM significantly outperforms strong VLM baselines, achieving state-of-the-art performance in operation semantics prediction. Furthermore, experiments in Android World show that our paradigm yields consistent Task Success Rate improvements for downstream agents. Together, Teach VLM and the Teach-and-Repeat paradigm offer a practical pathway from raw demonstrations to reusable task automation.

Problem

• mobile GUI understanding
• operational knowledge extraction
• vision-language models
• action comprehension
• UI automation

Innovation

• Teach-and-Repeat
• Vision-Language Model (VLM)
• Operational Knowledge Extraction
• Mobile GUI Automation
• Data Flywheel