File size: 15,070 Bytes
473a751 1be138d 473a751 1be138d 473a751 1be138d 473a751 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 | ---
language:
- en
- zh
license: other
pretty_name: ReflexBench Simulation Task Examples
configs:
- config_name: all
data_files:
- split: examples
path: data/all.parquet
default: true
- config_name: task_000000
data_files:
- split: examples
path: data/task_000000.parquet
- config_name: task_000001
data_files:
- split: examples
path: data/task_000001.parquet
- config_name: task_000002
data_files:
- split: examples
path: data/task_000002.parquet
- config_name: task_000003
data_files:
- split: examples
path: data/task_000003.parquet
- config_name: task_000004
data_files:
- split: examples
path: data/task_000004.parquet
- config_name: task_000005
data_files:
- split: examples
path: data/task_000005.parquet
- config_name: reaction_critical
data_files:
- split: examples
path: data/reaction_critical.parquet
tags:
- robotics
- video
- simulation
license_name: upstream-reflexbench-terms
license_link: LICENSE
---
# ReflexBench Simulation Task Examples
[打开 MP4 任务浏览页](https://travor278-reflexbench-task-examples.static.hf.space/?task=0)
Six official task_index values 0–5. Single Franka arm; only fixed_cam and wrist_cam exist. Do not relabel the wrist as left/right. Success rules describe evaluation, not an unprovided per-demo score.
所有示例是完整 H.264 MP4。单臂相机保留官方名称：fixed_cam / wrist_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 | Ball Catching | Reaction-critical | [打开](https://travor278-reflexbench-task-examples.static.hf.space/?task=0) |
| 1 | Conveyor Belt Pick And Place | Reaction-critical | [打开](https://travor278-reflexbench-task-examples.static.hf.space/?task=1) |
| 2 | Whack A Mole | Reaction-critical | [打开](https://travor278-reflexbench-task-examples.static.hf.space/?task=2) |
| 3 | Rolling Ball Interception | Reaction-critical | [打开](https://travor278-reflexbench-task-examples.static.hf.space/?task=3) |
| 4 | Ball Throwing | Reaction-critical | [打开](https://travor278-reflexbench-task-examples.static.hf.space/?task=4) |
| 5 | Rotating Peg Insertion | Reaction-critical | [打开](https://travor278-reflexbench-task-examples.static.hf.space/?task=5) |
ReflexBench 上游 episode 本身很短，多数选取记录只有约 1 秒。此处保持完整官方 length 和时间区间，没有人为截短、慢放或补帧。
## task_index=0 · Ball Catching
**Instruction:** hold the container with the gripper and catch the thrown ball in the container
**Description:** Catch a launched ball with a held container.
**Success / evaluation:** Episode success is task_phase == 4. The ball must remain in the virtual catch zone below the end effector for the configured consecutive-step threshold.
[官方说明](https://github.com/LxRoboticsLab/ReflexBench/blob/8bb931485093c6d98f8729774ad01bf824964e16/source/reflexbench/reflexbench/tasks/manager_based/ball_catching/README.md)
**Gym ID:** `BallCatching-Franka-DataCollection-v0`
**episode_index=0** · 24 frames / 0.96 s
[同步 MP4](media/ball_catching/episode_000000/synchronized.mp4) · [fixed_cam MP4](media/ball_catching/episode_000000/fixed_cam.mp4) · [wrist_cam MP4](media/ball_catching/episode_000000/wrist_cam.mp4) · [官方轨迹](trajectories/ball_catching/episode_000000.parquet)
**episode_index=100** · 27 frames / 1.08 s
[同步 MP4](media/ball_catching/episode_000100/synchronized.mp4) · [fixed_cam MP4](media/ball_catching/episode_000100/fixed_cam.mp4) · [wrist_cam MP4](media/ball_catching/episode_000100/wrist_cam.mp4) · [官方轨迹](trajectories/ball_catching/episode_000100.parquet)
**episode_index=199** · 37 frames / 1.48 s
[同步 MP4](media/ball_catching/episode_000199/synchronized.mp4) · [fixed_cam MP4](media/ball_catching/episode_000199/fixed_cam.mp4) · [wrist_cam MP4](media/ball_catching/episode_000199/wrist_cam.mp4) · [官方轨迹](trajectories/ball_catching/episode_000199.parquet)
## task_index=1 · Conveyor Belt Pick And Place
**Instruction:** pick up the cube from the moving conveyor and place it into the bin
**Description:** Pick from the moving conveyor and release into the bin.
**Success / evaluation:** Success requires the object inside the bin in environment-local coordinates: x in [-0.2,0.2], y in [-0.6,-0.3], z in [0.0,0.08] m.
[官方说明](https://github.com/LxRoboticsLab/ReflexBench/blob/8bb931485093c6d98f8729774ad01bf824964e16/source/reflexbench/reflexbench/tasks/manager_based/conveyor_belt_pick_and_place/README.md)
**Gym ID:** `ConveyorBeltPickAndPlace-Franka-DataCollection-v0`
**episode_index=200** · 31 frames / 1.24 s
[同步 MP4](media/conveyor_belt_pick_and_place/episode_000200/synchronized.mp4) · [fixed_cam MP4](media/conveyor_belt_pick_and_place/episode_000200/fixed_cam.mp4) · [wrist_cam MP4](media/conveyor_belt_pick_and_place/episode_000200/wrist_cam.mp4) · [官方轨迹](trajectories/conveyor_belt_pick_and_place/episode_000200.parquet)
**episode_index=318** · 37 frames / 1.48 s
[同步 MP4](media/conveyor_belt_pick_and_place/episode_000318/synchronized.mp4) · [fixed_cam MP4](media/conveyor_belt_pick_and_place/episode_000318/fixed_cam.mp4) · [wrist_cam MP4](media/conveyor_belt_pick_and_place/episode_000318/wrist_cam.mp4) · [官方轨迹](trajectories/conveyor_belt_pick_and_place/episode_000318.parquet)
**episode_index=435** · 36 frames / 1.44 s
[同步 MP4](media/conveyor_belt_pick_and_place/episode_000435/synchronized.mp4) · [fixed_cam MP4](media/conveyor_belt_pick_and_place/episode_000435/fixed_cam.mp4) · [wrist_cam MP4](media/conveyor_belt_pick_and_place/episode_000435/wrist_cam.mp4) · [官方轨迹](trajectories/conveyor_belt_pick_and_place/episode_000435.parquet)
## task_index=2 · Whack A Mole
**Instruction:** hit the mole that pops up on the board with the closed gripper
**Description:** Strike the active mole using the closed gripper during its popup window.
**Success / evaluation:** Unified evaluation success is valid_hits > 0. A valid hit requires proximity to the active mole, closed gripper and the configured dwell. Current eval profile uses 2.0 s popup, 0.3 s gap, 0 s initial delay and 3.0 s evaluation length. Completing the window schedule alone is not success.
[官方说明](https://github.com/LxRoboticsLab/ReflexBench/blob/8bb931485093c6d98f8729774ad01bf824964e16/source/reflexbench/reflexbench/tasks/manager_based/whack_a_mole/README.md)
**Gym ID:** `WhackAMole-Franka-DataCollection-v0`
**episode_index=436** · 18 frames / 0.72 s
[同步 MP4](media/whack_a_mole/episode_000436/synchronized.mp4) · [fixed_cam MP4](media/whack_a_mole/episode_000436/fixed_cam.mp4) · [wrist_cam MP4](media/whack_a_mole/episode_000436/wrist_cam.mp4) · [官方轨迹](trajectories/whack_a_mole/episode_000436.parquet)
**episode_index=536** · 16 frames / 0.64 s
[同步 MP4](media/whack_a_mole/episode_000536/synchronized.mp4) · [fixed_cam MP4](media/whack_a_mole/episode_000536/fixed_cam.mp4) · [wrist_cam MP4](media/whack_a_mole/episode_000536/wrist_cam.mp4) · [官方轨迹](trajectories/whack_a_mole/episode_000536.parquet)
**episode_index=635** · 16 frames / 0.64 s
[同步 MP4](media/whack_a_mole/episode_000635/synchronized.mp4) · [fixed_cam MP4](media/whack_a_mole/episode_000635/fixed_cam.mp4) · [wrist_cam MP4](media/whack_a_mole/episode_000635/wrist_cam.mp4) · [官方轨迹](trajectories/whack_a_mole/episode_000635.parquet)
## task_index=3 · Rolling Ball Interception
**Instruction:** hold the container with the gripper and catch the orange rolling down the ramp in the container
**Description:** Intercept the ball or orange rolling down the ramp.
**Success / evaluation:** Episode success is task_phase == 4. The rolling object must stay in the virtual catch zone for the configured hold threshold.
[官方说明](https://github.com/LxRoboticsLab/ReflexBench/blob/8bb931485093c6d98f8729774ad01bf824964e16/source/reflexbench/reflexbench/tasks/manager_based/rolling_ball_interception/README.md)
**Gym ID:** `RollingBallInterception-Franka-DataCollection-v0`
**episode_index=636** · 26 frames / 1.04 s
[同步 MP4](media/rolling_ball_interception/episode_000636/synchronized.mp4) · [fixed_cam MP4](media/rolling_ball_interception/episode_000636/fixed_cam.mp4) · [wrist_cam MP4](media/rolling_ball_interception/episode_000636/wrist_cam.mp4) · [官方轨迹](trajectories/rolling_ball_interception/episode_000636.parquet)
**episode_index=736** · 26 frames / 1.04 s
[同步 MP4](media/rolling_ball_interception/episode_000736/synchronized.mp4) · [fixed_cam MP4](media/rolling_ball_interception/episode_000736/fixed_cam.mp4) · [wrist_cam MP4](media/rolling_ball_interception/episode_000736/wrist_cam.mp4) · [官方轨迹](trajectories/rolling_ball_interception/episode_000736.parquet)
**episode_index=835** · 26 frames / 1.04 s
[同步 MP4](media/rolling_ball_interception/episode_000835/synchronized.mp4) · [fixed_cam MP4](media/rolling_ball_interception/episode_000835/fixed_cam.mp4) · [wrist_cam MP4](media/rolling_ball_interception/episode_000835/wrist_cam.mp4) · [官方轨迹](trajectories/rolling_ball_interception/episode_000835.parquet)
## task_index=4 · Ball Throwing
**Instruction:** toss the ball into the bin
**Description:** Throw the pre-grasped ball into the target bin.
**Success / evaluation:** Success is task_phase == 2, meaning the ball entered the target bin. Documentation bounds relative to bin center: |x-cx| <= 0.13, |y-cy| <= 0.09, z in [0.02,0.18] m.
[官方说明](https://github.com/LxRoboticsLab/ReflexBench/blob/8bb931485093c6d98f8729774ad01bf824964e16/source/reflexbench/reflexbench/tasks/manager_based/ball_throwing/README.md)
**Gym ID:** `BallThrowing-Franka-DataCollection-v0`
**episode_index=836** · 33 frames / 1.32 s
[同步 MP4](media/ball_throwing/episode_000836/synchronized.mp4) · [fixed_cam MP4](media/ball_throwing/episode_000836/fixed_cam.mp4) · [wrist_cam MP4](media/ball_throwing/episode_000836/wrist_cam.mp4) · [官方轨迹](trajectories/ball_throwing/episode_000836.parquet)
**episode_index=936** · 35 frames / 1.40 s
[同步 MP4](media/ball_throwing/episode_000936/synchronized.mp4) · [fixed_cam MP4](media/ball_throwing/episode_000936/fixed_cam.mp4) · [wrist_cam MP4](media/ball_throwing/episode_000936/wrist_cam.mp4) · [官方轨迹](trajectories/ball_throwing/episode_000936.parquet)
**episode_index=1035** · 33 frames / 1.32 s
[同步 MP4](media/ball_throwing/episode_001035/synchronized.mp4) · [fixed_cam MP4](media/ball_throwing/episode_001035/fixed_cam.mp4) · [wrist_cam MP4](media/ball_throwing/episode_001035/wrist_cam.mp4) · [官方轨迹](trajectories/ball_throwing/episode_001035.parquet)
## task_index=5 · Rotating Peg Insertion
**Instruction:** insert the peg into the hole on the rotating disc
**Description:** Time peg insertion into the moving hole of a rotating disc.
**Success / evaluation:** Success is task_phase == 4. Peg tip enters the hole within its radius tolerance, descends at least 0.02 m below the disc surface, and holds for at least 5 control steps (task README).
[官方说明](https://github.com/LxRoboticsLab/ReflexBench/blob/8bb931485093c6d98f8729774ad01bf824964e16/source/reflexbench/reflexbench/tasks/manager_based/rotating_peg_insertion/README.md)
**Gym ID:** `RotatingPegInsertion-Franka-DataCollection-v0`
**episode_index=1036** · 28 frames / 1.12 s
[同步 MP4](media/rotating_peg_insertion/episode_001036/synchronized.mp4) · [fixed_cam MP4](media/rotating_peg_insertion/episode_001036/fixed_cam.mp4) · [wrist_cam MP4](media/rotating_peg_insertion/episode_001036/wrist_cam.mp4) · [官方轨迹](trajectories/rotating_peg_insertion/episode_001036.parquet)
**episode_index=1136** · 59 frames / 2.36 s
[同步 MP4](media/rotating_peg_insertion/episode_001136/synchronized.mp4) · [fixed_cam MP4](media/rotating_peg_insertion/episode_001136/fixed_cam.mp4) · [wrist_cam MP4](media/rotating_peg_insertion/episode_001136/wrist_cam.mp4) · [官方轨迹](trajectories/rotating_peg_insertion/episode_001136.parquet)
**episode_index=1235** · 34 frames / 1.36 s
[同步 MP4](media/rotating_peg_insertion/episode_001235/synchronized.mp4) · [fixed_cam MP4](media/rotating_peg_insertion/episode_001235/fixed_cam.mp4) · [wrist_cam MP4](media/rotating_peg_insertion/episode_001235/wrist_cam.mp4) · [官方轨迹](trajectories/rotating_peg_insertion/episode_001235.parquet)
## 来源、许可与复现
官方数据：[原仓库](https://huggingface.co/datasets/cyx337/ReflexBench_dataset) · 固定 revision `9295b6e9878609a992047f0b8b65421a493299e7`。
官方代码：[GitHub](https://github.com/LxRoboticsLab/ReflexBench) · revision `8bb931485093c6d98f8729774ad01bf824964e16`。
保留上游 LICENSE 和来源文件在 `reference/`。ReflexBench 官方数据未单列许可证元数据；保留官方实现 BSD-3-Clause 来源，不扩大作者权利。
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`。
|