Qwen-Planner-Agent: A Closed-Loop AI-for-AI Framework for Real-World Mobile Planner Agents
Qwen-Planner-Agent uses a closed-loop AI-for-AI pipeline and leads MobilePA-Bench for mobile planning.
Qwen-Planner-Agent is developed in a closed-loop AI-for-AI framework aimed at long-horizon mobile planning, where real-device interaction is costly. Specialized agents generate tasks and trajectories and use training feedback to steer later data, while training combines a supervised planning cold start with hybrid-environment online reinforcement learning and Competence-Aware Reward-and-Advantage Engineering (CARE). An execution loop co-evolves memory, skills, and tools and feeds failure traces back into model and harness updates. The system is reported best overall on MobilePA-Bench, improving tool use, memory, skills, and sub-agent coordination over its base model, with gains on non-mobile agent benchmarks and general capabilities largely preserved.
- A shared action-feedback-verification contract links data, training, and deployment.
- Agents build tasks, collect trajectories, and curate data under human gating.
- CARE reduces reasoning and tool-use cost while preserving task performance.
- It leads MobilePA-Bench and improves non-mobile agent benchmarks over its base model.
Full article242 words · extracted from huggingface.co · click to collapse
The rapid progression of large language models is extending AI from passive content generation into the active workflows of engineering and scientific discovery. This shift raises a compelling question: can AI be both the object of development and an active participant in building next-generation AI systems? We explore this question by building Qwen-Planner-Agent within a closed-loop AI-for-AI framework for scalable development and iterative improvement. Mobile planning offers a demanding test of this approach: complex, long-horizon tasks challenge agent reliability, while costly real-device interaction limits development scalability. The framework connects data production, model training, and deployment through a shared action-feedback-verification contract. (i) AI for Data builds a human-gated agentic data flywheel in which specialized agents construct tasks, collect interaction trajectories, curate and balance training data, and use training feedback to guide subsequent data generation. (ii) AI for Training combines a supervised planning cold start with hybrid-environment online agentic reinforcement learning, where we introduce Competence-Aware Reward-and-Advantage Engineering (CARE) to reduce reasoning and tool-use costs while preserving task performance. (iii) AI drives model--harness co-evolution through an execution-evidence-driven loop that orchestrates memory, skills, and tools at runtime and feeds structured action feedback and preserved failure traces back into coordinated model and harness adaptation. Qwen-Planner-Agent achieves the best overall performance among all evaluated models and systems on MobilePA-Bench, improving over its base model across tool use, memory, skills, and sub-agent coordination. Further evaluations of our model show improvements across non-mobile agentic benchmarks while largely preserving general capabilities.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.29892