so101_cup_to_bluecircle_v1 — SO-101 cup repositioning, 50 episodes

▶ Full collection video (YouTube): https://youtu.be/s7c5Ls-Nkd0

Top-view (front camera), 8× speed — first 9 episodes, each starting the cup at a different position.

50 teleoperated demonstrations on the SO-101 arm for the task:

"pick up the cup near the blue circle and place the cup to the blue circle"

This dataset is part of a larger coffee-making robot project. The task is the cup-repositioning primitive, recorded in the project's actual workspace — the coffee machines stay in frame on purpose: this is the scene the policy is meant to run in.

A companion set with 20 additional, more varied placements: nevertmr/so101_cup_to_bluecircle_v1_var

This dataset was created using LeRobot.

Quick facts

Episodes / frames	50 / 19,138 @ 30 fps
Episode length	10.2 – 16.6 s (avg 12.8 s)
Cameras	front (fixed top-down workspace view), wrist (gripper-mounted) — both 640×480
Robot	SO-101 follower, teleoperated with an SO-101 leader arm
Recorded with	official LeRobot v0.4.4 CLI, on an Apple Silicon MacBook — no external compute
Format	LeRobotDataset v3.0

Scene geometry

• Two fixed cameras — positions never changed during collection, so pixel coordinates are consistent across all episodes.
• The blue circle is fixed to the table. The cup is the only object that moves between episodes. (In the _var companion set the circle also moves — see its card.)
• Background coffee machines are the deployment context of the parent project — kept exactly as-is in every frame.

Placement design

• The cup starts at a different position in every episode, spread across roughly eight directions around the circle at varied close-range distances. The goal is directional diversity, not a strict grid.
• Far placements are deliberately excluded — that range is held out as a controlled out-of-distribution condition for downstream experiments (e.g., failure detection). If you train on this data, expect degradation once the cup starts well outside the collected range. That's by design.

Placement coverage

All 50 cup start positions, extracted from a 4K top-view recording of the session (one frame at the start of each episode) and plotted on the shared background:

Per-episode start frames, numbered 0–49:

Recording protocol (identical in every episode)

1. Fixed home pose, gripper open.
2. One second of stillness at the start — frame 0 gives a clean view of the cup's start position, so displacement can be auto-labeled. (Measured: max joint std in the first second ≤ 0.15°.)
3. Approach → single grasp at mid-height, no regrasping → lift → transfer.
4. Release only after the cup is in full contact with the table inside the circle → retreat up and back.
5. One second of stillness at the end — the final still second is ground truth for the success state.

Fumbled or regrasped episodes were re-recorded on the spot, never patched afterwards.

Usage

from lerobot.datasets.lerobot_dataset import LeRobotDataset

ds = LeRobotDataset("nevertmr/so101_cup_to_bluecircle_v1")
print(ds.num_episodes, ds.num_frames)

Versioning

v1 — initial release (July 2026). 50 episodes, close-range placements only. Future re-collections (wider placements, more episodes, protocol changes) will be released as new versioned repos rather than overwriting this one.

Limitations

• Single operator, single session — lighting constant throughout.
• One task instruction on every frame, character-for-character identical.
• Joint values are calibrated degrees (per-device calibration), not raw motor ticks.