MobileAgent: enhancing mobile control via human-machine interaction and SOP integration
Tinghe Ding
TL;DR
This work tackles privacy and data-exploration challenges in LLM-driven mobile control by integrating user-specific Standard Operating Procedures into in-context learning and enabling interactive human-in-the-loop alignment. It introduces a probabilistic, context-aware decision framework for mobile agents and demonstrates how SOP information can be incorporated without increasing inference costs. Evaluated on the AitW benchmark with 30K multi-step instructions, the SOP-enhanced agent achieves a 66.92% overall action success, setting a new performance bar for this domain. The authors release code and data on GitHub to support reproducibility and further research.
Abstract
Agents centered around Large Language Models (LLMs) are now capable of automating mobile device operations for users. After fine-tuning to learn a user's mobile operations, these agents can adhere to high-level user instructions online. They execute tasks such as goal decomposition, sequencing of sub-goals, and interactive environmental exploration, until the final objective is achieved. However, privacy concerns related to personalized user data arise during mobile operations, requiring user confirmation. Moreover, users' real-world operations are exploratory, with action data being complex and redundant, posing challenges for agent learning. To address these issues, in our practical application, we have designed interactive tasks between agents and humans to identify sensitive information and align with personalized user needs. Additionally, we integrated Standard Operating Procedure (SOP) information within the model's in-context learning to enhance the agent's comprehension of complex task execution. Our approach is evaluated on the new device control benchmark AitW, which encompasses 30K unique instructions across multi-step tasks, including application operation, web searching, and web shopping. Experimental results show that the SOP-based agent achieves state-of-the-art performance in LLMs without incurring additional inference costs, boasting an overall action success rate of 66.92\%. The code and data examples are available at https://github.com/alipay/mobile-agent.