ARCANA: A Reflective Multi-Agent Program Synthesis Framework for ARC-AGI-2 Reasoning

ARCANA：用于 ARC-AGI-2 推理的反射式多智能体程序合成框架

Abstract: We present ARCANA, a collaborative multi-agent framework for solving ARC-AGI-2 tasks under strict test time and hardware constraints. 摘要： 我们提出了 ARCANA，这是一个协作式多智能体框架，旨在解决在严格的测试时间和硬件约束下的 ARC-AGI-2 任务。

ARCANA decomposes each task into iterative perception, hypothesis generation, symbolic execution, and reflective refinement. ARCANA 将每个任务分解为迭代感知、假设生成、符号执行和反射式优化四个阶段。

A perceptual grounding agent builds object-centric scene graphs from raw grids, a latent program policy proposes diverse DSL programs, a symbolic executor verifies candidates on demonstrations, and a reflective agent synthesizes failure-driven feedback for the next turn. 其中，感知基础智能体（perceptual grounding agent）从原始网格中构建以对象为中心的场景图，潜在程序策略（latent program policy）提出多样化的 DSL 程序，符号执行器（symbolic executor）在演示案例上验证候选程序，而反射智能体（reflective agent）则为下一轮迭代合成基于失败驱动的反馈。

These agents communicate through a shared differentiable blackboard and are scheduled by a learned meta-controller. 这些智能体通过一个共享的可微黑板进行通信，并由一个学习型的元控制器（meta-controller）进行调度。

The design combines structured program search with adaptive multi-turn correction, improving reasoning efficiency and solution quality on challenging abstract transformation tasks. 该设计将结构化程序搜索与自适应多轮修正相结合，提高了在具有挑战性的抽象转换任务中的推理效率和解决方案质量。