视觉-语言-动作模型的分位数头
Quantile Head for Vision-Language-Action Models
- Tsinghua University（清华大学）
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
提出分位数头，统一回归与流匹配，通过联合监督中位数策略，在LIBERO及真实机器人任务上取得最高成功率与最短时间。
AI中文摘要：
视觉-语言-动作（VLA）模型将预训练的视觉-语言模型（VLMs）与动作头集成，用于机器人控制。常见的动作头存在明显局限：点回归仅提供动作分布的点估计，而标准流匹配采样器需要昂贵的迭代采样。为解决这些局限，我们将回归和流匹配统一在共享目标下，并扩展以推导出分位数目标。该分位数目标指导了我们的分位数头（Quantile Head）的设计，它预测中位数和正间隔，在一次前向传播中形成有序的边缘动作分位数。这些分位数支持多种采样策略而无需重新训练，并联合监督以训练默认的中位数策略。我们对这种联合监督的局部分析表明，在附近分位数校准、固定间隔和匹配的修正速度下，直接中位数更新的方差低于仅中位数监督下的方差。实验表明，这种联合监督的中位数策略在LIBERO、LIBERO-Plus、LIBERO-Pro和两个真实机器人任务上，与所比较的方法相比取得了最高的平均成功率，并且在匹配的LIBERO基线中具有最短的平均情节时间；代码可在该HTTPS URL获取。
英文摘要：
Vision-Language-Action (VLA) models integrate pretrained Vision-Language Models (VLMs) with action heads for robot control. Common action heads have distinct limitations: point regression provides only a point estimate of the action distribution, while standard flow-matching samplers require costly iterative sampling. To address these limitations, we unify regression and flow matching under a shared objective and extend it to derive a quantile objective. This quantile objective guides the design of our Quantile Head, which predicts a median and positive gaps to form ordered marginal action quantiles in one forward pass. These quantiles support multiple sampling strategies without retraining and are jointly supervised to train the default median policy. Our local analysis of this joint supervision shows that, with calibrated nearby quantiles, fixed gaps, and matched correction speed, direct median updates have lower variance than under median-only supervision. Experiments show that this jointly supervised median policy achieves the highest average success rates among the compared methods on LIBERO, LIBERO-Plus, LIBERO-Pro, and two real-robot tasks, together with the shortest mean episode time among matched LIBERO baselines; code is available at https://github.com/xwangrs/Quantile-Head-for-VLA.
