RLinf/rlt-maniskill-PegInsertionSide-v1-400-succ
收藏资源简介:
rlt_maniskill_joint是一个LeRobot风格的数据集,用于在ManiSkill插孔插入任务上进行关节控制机器人学习令牌(RLT)训练。它专为RLinf + OpenPI的pi05_rlt_joint流水线设计,并用于三个阶段:1. OpenPI监督微调(SFT)基础策略训练;2. RLT第1阶段RL令牌训练;3. RLT第2阶段在线RL初始化和归一化。该数据集对应ManiSkill任务:环境为PegInsertionSideWideClearance-v1,控制模式为pd_joint_delta_pos,默认指令为insert the peg in the hole。此数据集旨在用于关节空间视觉语言动作训练,而非末端执行器动作预测。
`rlt_maniskill_joint` is a LeRobot-style dataset for joint-control Robot Learning Token (RLT) training on the ManiSkill peg insertion task. It is designed for the RLinf + OpenPI `pi05_rlt_joint` pipeline and is used in three stages: 1. OpenPI supervised fine-tuning (SFT) base policy training 2. RLT Stage 1 RL-token training 3. RLT Stage 2 online RL initialization and normalization The dataset corresponds to the ManiSkill task: - Environment: `PegInsertionSideWideClearance-v1` - Control mode: `pd_joint_delta_pos` - Default instruction: `insert the peg in the hole` This dataset is intended for joint-space vision-language-action training rather than end-effector action prediction.




