关于VLA微调的阶段信息接口的实证研究
An Empirical Study on Stage-Information Interfaces for VLA Fine-Tuning
浏览论文内容
中文总结 AI 辅助
研究长期操作中VLA微调的阶段信息接口，通过分段动作注释等方法，在直接微调与延续微调下和GR00T N1.6比较，发现不同接口表示和训练安排下阶段信息对策略成功率影响不同。
中文摘要 AI 辅助
在长期操作中，一个高级指令可涵盖多个动作阶段。我们使用分段动作注释作为全任务指令和VLA动作块之间的中间表示。一个进度模块跟踪活动阶段，而动作策略接收阶段信息，其形式可以是当前阶段文本或机器人状态中的归一化序数阶段索引。我们在直接微调以及从全任务指令基线进行的延续微调下，将这些接口与LIBERO - 10上的GR00T N1.6进行比较。直接微调时，全任务指令、当前阶段文本和序数阶段状态的平均成功率分别为57.45%、50.24%和54.36%，表明明确的阶段信息不会自动改善策略。延续微调时，相应均值为49.07%、50.00%和53.75%，序数阶段状态在所有三次配对运行中均超过其他两者。观察到的益处因接口表示和训练安排而异。
英文摘要
One high-level instruction in long-horizon manipulation can cover several action stages. We use segmented action annotations as an intermediate representation between the full-task instruction and VLA action chunks. A progress module tracks the active stage, while the action policy receives stage information either as current-stage text or as a normalized ordinal stage index in robot state. We compare these interfaces with GR00T N1.6 on LIBERO-10 under direct fine-tuning and continuation fine-tuning from a full-task instruction baseline. Under direct fine-tuning, full-task instruction, current-stage text, and Ordinal Stage-State achieve mean success rates of 57.45%, 50.24%, and 54.36%, respectively, showing that explicit stage information does not automatically improve the policy. Under continuation, the corresponding means are 49.07%, 50.00%, and 53.75%, with Ordinal Stage-State exceeding both alternatives in all three paired runs. The observed benefit differs across interface representations and training arrangements.
