single-view
libero-singleview-idm
LIBERO 单视角离线 IDM：Codex 技术交接与运行手册
本文档是项目迁移到另一台机器后，供 Codex 或工程人员直接接手使用的
主运行手册。除非明确需要改变实验设计，否则应以本文、cfg/ 和
scripts/ 中的配置为准。
0. 接手结论与执行原则
项目目标是把离线 LIBERO 单视角视频转换为机器人动作：
agentview 离线视频
→ 每 16 帧生成 20×20 个均匀点轨迹
→ 单视角 Motion Tokenizer
→ 对应 LIBERO 系列的 IDM
→ 16 步 × 7 维动作
接手此项目的 Codex 应遵守以下原则：
先进入项目根目录，再执行任何脚本；所有默认路径均相对于项目根目录。
先运行资产校验。校验通过时，不要重新下载或重新生成 LIBERO 和轨迹数据。
先进行单卡或 DDP 冒烟测试，再启动完整训练。
Motion Tokenizer 只训练一个共享模型；五个 LIBERO 系列分别训练自己的 IDM。
不训练 Forward… See the full description on the dataset page: https://huggingface.co/datasets/yimmmmmy/libero-singleview-idm.bridge_single_view
bridge_single_view
This dataset was created using LeRobot.
Dataset Description
bridge_single_view is a full-data repack of Bridge v2 into LeRobot v3 with a
single canonical visual stream:
all source episodes and tasks are preserved
only the always-present primary camera is kept
observation.images.image_0 is renamed to observation.images.image
camera_present is removed because it is degenerate in the single-view output
Dataset Structure… See the full description on the dataset page: https://huggingface.co/datasets/chaoqi-liu/bridge_single_view.sweep_80demos_singleview_20260830_155934This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"action": {
"dtype": "float32",
"names": [
"shoulder_pan.pos",
"shoulder_lift.pos",
"elbow_flex.pos",
"wrist_flex.pos",
"wrist_roll.pos",
"gripper.pos"
],
"shape": [
6… See the full description on the dataset page: https://huggingface.co/datasets/saipuneethgottam/sweep_80demos_singleview_20260830_155934.stacking_80demos_singleviewThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"action": {
"dtype": "float32",
"names": [
"shoulder_pan.pos",
"shoulder_lift.pos",
"elbow_flex.pos",
"wrist_flex.pos",
"wrist_roll.pos",
"gripper.pos"
],
"shape": [
6… See the full description on the dataset page: https://huggingface.co/datasets/saipuneethgottam/stacking_80demos_singleview.pickplace_80demos_singleviewThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"action": {
"dtype": "float32",
"names": [
"shoulder_pan.pos",
"shoulder_lift.pos",
"elbow_flex.pos",
"wrist_flex.pos",
"wrist_roll.pos",
"gripper.pos"
],
"shape": [
6… See the full description on the dataset page: https://huggingface.co/datasets/saipuneethgottam/pickplace_80demos_singleview.rlvr-code-view-single-turn-new