5hadytru/so101_bench_sim_4

SO101 Bench Sim 4 — Sim-Real Correspondence Dataset

so101_bench_sim_4 is a curated simulated robot-manipulation dataset for the SO-101 Bench sim-real correspondence study. Its tabletop object pool is the SO-101 Bench real-world “seen” object set: the simulated objects are the counterparts of the objects used in the study's real-world seen condition. The dataset is intended to support controlled comparisons between simulation and real observations/actions for that shared object set.

Composition

sim_4 is a LeRobot v3.0 aggregate, in this order:

1. so101_bench_sim_2: 398 episodes and 128,784 frames.
2. so101_bench_sim_3: 487 curated episodes and 161,514 frames.

The source task vocabularies contain 263 and 323 instructions respectively; 52 instructions overlap, yielding 534 distinct task descriptions in the merged dataset. All 534 are represented in the data.

Property	Value
Format	LeRobot v3.0
Robot type	so101_follower
Environment	SO101 Bench simulated tabletop bin manipulation
Split	train only (0:885)
Episodes	885
Frames	290,298
Frame rate	30 FPS
Total recorded duration	9,676.6 s (2 h 41 min 16.6 s)
Episode length	143–1,166 frames; 328.02 frames on average
Distinct language tasks	534

This is a simulation dataset: it does not contain real-camera recordings or real-robot trajectories. The object-set correspondence is deliberate, but sim-to-real transfer and performance on physical hardware must be evaluated separately.

Observations and actions

Each time step contains:

• observation.state: six float32 robot position values in this order: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, and gripper.pos.
• action: six float32 action-position values in the same order.
• observation.images.front: 640 × 480 RGB front-camera video.
• observation.images.overhead: 640 × 480 RGB overhead-camera video.
• timestamp, frame_index, episode_index, index, and task_index.

The camera streams are AV1, yuv420p, 30 FPS video without audio. In the v3.0 layout, videos are stored as multi-episode shards and the episode metadata identifies every episode's video span. meta/tasks.parquet maps task_index to the natural-language instruction.

Loading

This is a LeRobot v3.0 dataset. Use a LeRobot version that supports v3.0:

from lerobot.datasets.lerobot_dataset import LeRobotDataset

dataset = LeRobotDataset("5hadytru/so101_bench_sim_4")

Intended use and limitations

Use this data for sim-real correspondence analysis, simulation-based policy learning, and vision-language-action experiments involving the real-world seen SO-101 Bench object set. The delivered dataset has a training split only; it does not include rewards, per-episode success labels, or a separate validation/test split.