Travor278/DynamicVLA-Task-Examples

DynamicVLA Simulation Task Examples

打开 MP4 任务浏览页

DOM has 3 task families, not 3 task_index values: 140,932 structured instruction IDs and 207,306 episodes. This subset selects 6 demonstrations per family (18 total), preserving actual task_index/episode_index and structured instructions. All 3 published cameras are single-arm cameras: opposite, wrist, side, not top/left/right arms.

所有示例是完整 H.264 MP4。单臂相机保留官方名称：opst_cam / wrist_cam / side_cam。每条有同步合并视频、各路原视频与官方动作/状态 Parquet。

评估标准根据官方代码和文档整理；发布数据未保存逐条示教 success 标签，故 actual_episode_success 为 null。本文不编造 0–100 得分。

官方 task_index	Task / instruction scenario	分类	直接浏览
0	Long Horizon · white box	Long-Horizon	打开
70	Long Horizon · white box	Long-Horizon	打开
80	Long Horizon · box with Tencent Logo	Long-Horizon	打开
135	Long Horizon · plastic box	Long-Horizon	打开
196	Long Horizon · box	Long-Horizon	打开
246	Long Horizon · box	Long-Horizon	打开
10235	Pick · the moving object with the lower initial velocity	Pick	打开
10588	Pick · the object in the middle from left to right at the start	Pick	打开
42175	Pick · the object moving in the robot's forward-left direction	Pick	打开
42560	Pick · lime	Pick	打开
60569	Pick · the object that is closest to the robot's left at the start	Pick	打开
60993	Pick · the object that is closest to the robot's left at the start	Pick	打开
77322	Place · moving-forward object	Place	打开
77667	Place · beer bottle	Place	打开
108733	Place · lemon	Place	打开
111207	Place · lemon	Place	打开
140471	Place · black cylinder water bottle	Place	打开
140594	Place · water bottle	Place	打开

task_index=0 · Long Horizon · white box

Instruction: Pick up the entire set of objects and place them in/on the white box.

Description: Official DOM simulation scenario long-horizon_franka_apple00d_O03_00772319_1dc5-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "long_horizon",
  "objects": [
    "entire set of objects"
  ],
  "containers": [
    "white box",
    "box with Tesla Logo",
    "white box with Tesla Logo",
    "plastic box",
    "box"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=0 · 194 frames / 7.76 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=135 · Long Horizon · plastic box

Instruction: Pick up the entire set of objects and place them in/on the plastic box.

Description: Official DOM simulation scenario long-horizon_franka_apple08d_O03_00819030_04bc-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "long_horizon",
  "objects": [
    "entire set of objects"
  ],
  "containers": [
    "plastic box",
    "pink box",
    "pink plastic box",
    "box"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=200 · 227 frames / 9.08 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=196 · Long Horizon · box

Instruction: Pick up the entire set of objects and place them in/on the box.

Description: Official DOM simulation scenario long-horizon_franka_apple14d_O03_00888109_ee3b-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "long_horizon",
  "objects": [
    "entire set of objects"
  ],
  "containers": [
    "box",
    "pink box",
    "pink plastic box",
    "plastic box"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=400 · 205 frames / 8.20 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=246 · Long Horizon · box

Instruction: Pick up the fusiform avocado and place it in/on the box.

Description: Official DOM simulation scenario long-horizon_franka_avocado00d_O03_00769598_7aed-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "long_horizon",
  "objects": [
    "fusiform avocado",
    "the object that is closer to the robot's right at the start",
    "the object moving in the robot's forward-left direction",
    "green avocado",
    "green fusiform avocado",
    "avocado",
    "the moving object with the highest initial velocity"
  ],
  "containers": [
    "box",
    "green box",
    "plastic box",
    "green plastic box"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=600 · 220 frames / 8.80 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=80 · Long Horizon · box with Tencent Logo

Instruction: Pick up the entire set of objects and place them in/on the box with Tencent Logo.

Description: Official DOM simulation scenario long-horizon_franka_avocado04d_O03_00878095_8d75-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "long_horizon",
  "objects": [
    "entire set of objects"
  ],
  "containers": [
    "box with Tencent Logo",
    "plastic box",
    "box",
    "white box",
    "white box with Tencent Logo"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=800 · 232 frames / 9.28 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=70 · Long Horizon · white box

Instruction: Pick up the entire set of objects and place them in/on the white box.

Description: Official DOM simulation scenario long-horizon_franka_beer01d_O03_00794991_f96b-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "long_horizon",
  "objects": [
    "entire set of objects"
  ],
  "containers": [
    "white box",
    "white box with Tencent Logo",
    "box",
    "box with Tencent Logo",
    "plastic box"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=1000 · 209 frames / 8.36 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=10235 · Pick · the moving object with the lower initial velocity

Instruction: Pick up the moving object with the lower initial velocity.

Description: Official DOM simulation scenario pick_franka_avocado01d_O03_01140437_873f-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires object-to-end-effector distance < tolerance and end-effector-to-configured-goal distance < tolerance. The current default tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "pick",
  "objects": [
    "the moving object with the lower initial velocity",
    "green avocado",
    "avocado",
    "green long fusiform avocado",
    "long fusiform avocado",
    "the object that is closer to the robot's right at the start"
  ],
  "containers": [
    "tray",
    "white tray with CapitaLand logo",
    "white tray",
    "tray with CapitaLand logo"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=16460 · 83 frames / 3.32 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=10588 · Pick · the object in the middle from left to right at the start

Instruction: Pick up the object in the middle from left to right at the start.

Description: Official DOM simulation scenario pick_franka_avocado02d_O04_01042912_e0d2-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires object-to-end-effector distance < tolerance and end-effector-to-configured-goal distance < tolerance. The current default tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "pick",
  "objects": [
    "the object in the middle from left to right at the start",
    "green avocado",
    "avocado",
    "pear-shaped long avocado",
    "green pear-shaped long avocado",
    "the object farthest from the robot mounting edge at the start"
  ],
  "containers": [
    "square placemat features a bold, stylized illustration of a tui",
    "Square placemat with a bold, stylized tui",
    "square placemat",
    "square placemat with yellow background"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=16837 · 89 frames / 3.56 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=42175 · Pick · the object moving in the robot's forward-left direction

Instruction: Pick up the object moving in the robot's forward-left direction.

Description: Official DOM simulation scenario pick_franka_lime00d_O03_01174264_30d7-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires object-to-end-effector distance < tolerance and end-effector-to-configured-goal distance < tolerance. The current default tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "pick",
  "objects": [
    "the object moving in the robot's forward-left direction",
    "dark green lime",
    "the object that is closer to the robot's right at the start",
    "the object closer to the robot mounting edge at the start",
    "lime"
  ],
  "containers": [
    "tray",
    "tray with Meta logo",
    "white tray",
    "white tray with Meta logo"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=51235 · 87 frames / 3.48 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=42560 · Pick · lime

Instruction: Pick up the lime.

Description: Official DOM simulation scenario pick_franka_lime00d_O05_01312751_5a1e-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires object-to-end-effector distance < tolerance and end-effector-to-configured-goal distance < tolerance. The current default tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "pick",
  "objects": [
    "lime",
    "dark green lime"
  ],
  "containers": [
    "the box with the larger volume",
    "the box closer to the robot at the start",
    "the box that is closer to the robot's right at the start",
    "the box with the larger area",
    "red box",
    "the box farther from the robot mounting edge at the start",
    "box with MMLab At NTU words",
    "plastic box",
    "the taller box",
    "red box with MMLab At NTU words",
    "box"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=51658 · 87 frames / 3.48 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=60569 · Pick · the object that is closest to the robot's left at the start

Instruction: Pick up the object that is closest to the robot's left at the start.

Description: Official DOM simulation scenario pick_franka_tangerine00d_O04_01057450_ee6d-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires object-to-end-effector distance < tolerance and end-effector-to-configured-goal distance < tolerance. The current default tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "pick",
  "objects": [
    "the object that is closest to the robot's left at the start",
    "tangerine"
  ],
  "containers": [
    "pink plastic box",
    "pink box",
    "box",
    "plastic box"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=71104 · 95 frames / 3.80 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=60993 · Pick · the object that is closest to the robot's left at the start

Instruction: Pick up the object that is closest to the robot's left at the start.

Description: Official DOM simulation scenario pick_franka_tangerine03d_O04_01097900_93cd-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires object-to-end-effector distance < tolerance and end-effector-to-configured-goal distance < tolerance. The current default tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "pick",
  "objects": [
    "the object that is closest to the robot's left at the start",
    "tangerine"
  ],
  "containers": [
    "tray",
    "white tray",
    "tray with NTU logo",
    "white tray with NTU logo"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=71578 · 85 frames / 3.40 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=77322 · Place · moving-forward object

Instruction: Pick up the moving-forward object and place it in/on the ceramic plate with blue floral patterns.

Description: Official DOM simulation scenario place_franka_beer03d_O02_00082428_f57d-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "place",
  "objects": [
    "moving-forward object",
    "light color beer bottle with Heineken carvings",
    "beer bottle with Heineken carvings",
    "beer bottle",
    "moving-backward-left object",
    "moving-backward-right object",
    "moving-right object",
    "light color beer bottle",
    "moving-backward object",
    "moving-left object"
  ],
  "containers": [
    "ceramic plate with blue floral patterns",
    "ceramic plate",
    "plate",
    "plate with blue floral patterns"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=98470 · 111 frames / 4.44 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=77667 · Place · beer bottle

Instruction: Pick up the beer bottle and place it in/on the placemat that is closer to the robot's left at the start.

Description: Official DOM simulation scenario place_franka_beer03d_O04_00340177_6e45-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "place",
  "objects": [
    "beer bottle",
    "the object that is closer to the robot's left at the start",
    "beer bottle with Heineken carvings",
    "light color beer bottle",
    "the moving object with the highest initial velocity",
    "light color beer bottle with Heineken carvings"
  ],
  "containers": [
    "the placemat that is closer to the robot's left at the start",
    "green square placemat",
    "the placemat closer to the robot mounting edge at the start",
    "square placemat"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=98947 · 117 frames / 4.68 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=108733 · Place · lemon

Instruction: Pick up the lemon and place it in/on the tray.

Description: Official DOM simulation scenario place_franka_lemon06d_O02_00026504_29ce-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "place",
  "objects": [
    "lemon"
  ],
  "containers": [
    "tray",
    "white tray with HSBC logo",
    "tray with HSBC logo",
    "white tray"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=153474 · 108 frames / 4.32 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=111207 · Place · lemon

Instruction: Pick up the lemon and place it in/on the placemat farther from the robot mounting edge at the start.

Description: Official DOM simulation scenario place_franka_lemon06d_O04_00340971_5a64-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "place",
  "objects": [
    "lemon",
    "the object that is closer to the robot's right at the start",
    "the object moving in the robot's backward-left direction",
    "the moving object with the highest initial velocity"
  ],
  "containers": [
    "the placemat farther from the robot mounting edge at the start",
    "green square placemat"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=154278 · 110 frames / 4.40 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=140594 · Place · water bottle

Instruction: Pick up the water bottle and place it in/on the bowl.

Description: Official DOM simulation scenario place_franka_wbottle23d_O02_00046150_50a0-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "place",
  "objects": [
    "water bottle",
    "cylinder water bottle",
    "cylinder water bottle with grey and orange lid",
    "black cylinder water bottle",
    "black cylinder water bottle with grey and orange lid",
    "black water bottle",
    "black water bottle with grey and orange lid",
    "water bottle with grey and orange lid"
  ],
  "containers": [
    "bowl",
    "wooden bowl",
    "shallow bowl",
    "wooden shallow bowl"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=206737 · 124 frames / 4.96 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

task_index=140471 · Place · black cylinder water bottle

Instruction: Pick up the black cylinder water bottle and place it in/on the black box.

Description: Official DOM simulation scenario place_franka_wbottle23d_O02_00120780_166a-tr. Object/container aliases are preserved in the metadata.

Success / evaluation: Success requires ALL target objects to satisfy the container-relative geometric placement check, and the end effector to return within tolerance of the configured goal. Current default return tolerance is 0.015 m; per-scene overrides may differ.

官方说明

结构化原始 instruction（完整别名）：

{
  "task": "place",
  "objects": [
    "black cylinder water bottle",
    "cylinder water bottle with grey and orange lid",
    "black water bottle",
    "water bottle",
    "black cylinder water bottle with grey and orange lid",
    "cylinder water bottle",
    "black water bottle with grey and orange lid",
    "water bottle with grey and orange lid"
  ],
  "containers": [
    "black box",
    "box",
    "plastic box",
    "black plastic box"
  ]
}

根据官方结构化 instruction 和首个别名确定性呈现；原记录没有保存采集时使用的具体随机语言变体，完整别名见下方元数据。

episode_index=207108 · 105 frames / 4.20 s

同步 MP4 · opst_cam MP4 · wrist_cam MP4 · side_cam MP4 · 官方轨迹

来源、许可与复现

官方数据：原仓库 · 固定 revision bbf6baa15f7253746d271a6e958e037afa537b1e。

官方代码：GitHub · revision 434dc7105cf6d559018d15af5fecbcdad44f036e。

遵循上游 S-Lab License 1.0，仅允许非商业用途；见 LICENSE。

选择性提取记录、原始时间区间、task/episode 对应、相机元数据和校验均在 provenance/。不下载场景资产、模型权重或完整数据仓库。

复现脚本：scripts/build_examples.py。浏览页源码另见 browser/ 和 scripts/build_browser.py。