EXPO-FT 用 RL 后训练让 VLA 达到 95% 以上可靠性
How can VLAs achieve 95+% reliability?
AI 导读
EXPO-FT 通过 RL 后训练提升视觉语言动作模型（VLA）的可靠性，π0.5 在 8 个测试任务上均达到 30/30 成功率，平均仅使用 19 分钟 RL 数据。
查看原帖推出 EXPO-FT —— 高效、可靠且开源的 VLA 微调！ EXPO-FT 为 π0.5 解锁了具有挑战性的操作任务： 布置串灯并插入电源连接器使其点亮 将台球击入球袋 将花插入葡萄酒瓶 (1/5) https://t.co/wrEwsF211q
查看原文
Introducing EXPO-FT – Efficient, Reliable & Open-Source VLA Finetuning! EXPO-FT unlocks π0.5 for challenging manipulation tasks: Routing string lights & inserting the power connector to illuminate them Striking pool ball into pocket Inserting flower into wine bottle (1/5) https://t.co/wrEwsF211q
正文
How can VLAs achieve 95+% reliability?
Using RL post-training with EXPO-FT: - π0.5 improves to 30/30 success on all 8 tasks tested - uses only 19 min of RL data on average