Do Multi-Agents Dream of Electric Screens? Achieving Perfect Accuracy on AndroidWorld Through Task Decomposition

2026-02-08

AI Summary

This work addresses the limitations of existing single-agent systems on the AndroidWorld benchmark, which struggle with complex tasks due to context contamination, undetected text input failures, and repetitive action loops. To overcome these challenges, we propose Minitap, a novel multi-agent system featuring a six-agent cognitive separation architecture. Minitap decomposes tasks, validates text inputs deterministically, employs metacognitive reasoning, eliminates redundant actions, and incorporates a post-execution verification mechanism coupled with loop detection to trigger adaptive policy switching. Evaluated on all 116 tasks in the AndroidWorld benchmark, Minitap achieves a 100% success rate—surpassing single-agent baselines by 21 percentage points and becoming the first system to exceed human performance, which stands at 80%. The implementation has been publicly released.

Technology Category

• Multiagent Systems: Adversarial Agents
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Cognitive Modeling & Cognitive Systems: Agent Architectures

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

We present Minitap, a multi-agent system that achieves 100% success on the AndroidWorld benchmark, the first to fully solve all 116 tasks and surpassing human performance (80%). We first analyze why single-agent architectures fail: context pollution from mixed reasoning traces, silent text input failures undetected by the agent, and repetitive action loops without escape. Minitap addresses each failure through targeted mechanisms: cognitive separation across six specialized agents, deterministic post-validation of text input against device state, and meta-cognitive reasoning that detects cycles and triggers strategy changes. Ablations show multi-agent decomposition contributes +21 points over single-agent baselines; verified execution adds +7 points; meta-cognition adds +9 points. We release Minitap as open-source software. https://github.com/minitap-ai/mobile-use

Problem

• multi-agent systems
• Android automation
• task decomposition
• mobile UI interaction
• reasoning failures

Innovation

• multi-agent system
• task decomposition
• meta-cognitive reasoning
• deterministic validation
• Android automation