[论文] SUN: Persistent Programs For Language-Grounded Control-to-Learning-to-...

论文概要

研究领域: ML 作者: Weiqi Wang, Zhi Li, Yudong Lei 发布时间: 2025-09-01 arXiv: 2509.00140

中文摘要

在长程操作中将基于模型的控制与学习策略联系起来一直存在一个隐性的分歧：控制执行指定目标，学习将该行为摊销为反应式策略，但现有协议丢弃了任务语义，导致奖励需要手工设计且行为偏离控制验证的结果。我们引入语义统一（SUN）程序，这是一种类型化可执行文件，其中几何和接触关系只定义一次，然后编译为对齐的模型预测控制（MPC）成本、满足谓词、RL奖励、转移守卫和诊断。我们的系统Kuafu由大型视觉语言系统驱动，从语言和场景语义自动合成SUN程序，通过MPC筛选可行性，并在训练阶段条件策略时保留语义。在九项任务中，Kuafu达到82.03%的宏观成功率，优于稀疏奖励（35.67%）和Stage-BC（24.75%）基线。在8192路规模下，它每小时生成人类远程操作成功轨迹时间的10.57倍。每个任务500条轨迹，Kuafu数据训练DP3策略达到46.0%的模拟成功率（替代方案为22.4%），在物理Franka和Kinova机器人上达到34.7%。这些结果表明，模拟筛选的任务语义可以有效地将控制摊销为鲁棒策略，无需演示或手动密集奖励，统一了符号规划和数据驱动执行。

原文摘要

Bridging model-based control and learned policies in long-horizon manipulation has harbored a silent disagreement: control executes specified objectives, learning amortizes that behavior into a reactive policy, yet existing protocols discard task semantics, leaving rewards hand-crafted and behavior drifting from what control verified. We introduce Semantically UNified (SUN) Programs, typed executables where geometric and contact relations are defined once and compiled into aligned Model Predictive Control (MPC) costs, satisfaction predicates, RL rewards, transition guards, and diagnostics. Our system, Kuafu, driven by large vision language systems, automatically synthesizes SUN Programs from language and scene语义, screens feasibility via MPC, and retains semantics while training stage-condi...

#论文 #arXiv #ML #小凯