←Back to feed
🧠 AI🟢 Bullish
K^2-Agent: Co-Evolving Know-What and Know-How for Hierarchical Mobile Device Control
arXiv – CS AI|Zhe Wu, Donglin Mo, Hongjin Lu, Junliang Xing, Jianheng Liu, Yuheng Jing, Kai Li, Kun Shao, Jianye Hao, Yuanchun Shi||1 views
🤖AI Summary
Researchers introduce K²-Agent, a hierarchical AI framework for mobile device control that separates 'know-what' and 'know-how' knowledge to achieve 76.1% success rate on AndroidWorld benchmark. The system uses a high-level reasoner for task planning and low-level executor for skill execution, showing strong generalization across different models and tasks.
Key Takeaways
- →K²-Agent achieves 76.1% success rate on challenging AndroidWorld benchmark using only raw screenshots and open-source models.
- →The framework separates declarative knowledge (knowing what) from procedural knowledge (knowing how) to mimic human-like cognition.
- →High-level reasoner uses Summarize-Reflect-Locate-Revise loop to refine task knowledge from single demonstrations.
- →Low-level executor employs curriculum-guided Group Relative Policy Optimization for autonomous skill learning.
- →System demonstrates dual generalization with knowledge transfer across models and competitive performance on unseen tasks.
#mobile-ai#device-control#hierarchical-learning#reinforcement-learning#ai-agents#automation#benchmark#android#machine-learning#cognitive-ai
Read Original →via arXiv – CS AI
Act on this with AI
Stay ahead of the market.
Connect your wallet to an AI agent. It reads balances, proposes swaps and bridges across 15 chains — you keep full control of your keys.
Related Articles