ZenO Sim-Teleop — Franka Panda Manipulation (Sample)

Teleoperated Franka Emika Panda manipulation trajectories collected in a MuJoCo simulation, packaged in the LeRobot v2.1 format with synchronized front and wrist camera video.

This repository is a curated public sample. It contains 50 episodes drawn from 5 expert (long-horizon, contact-rich) tasks — 10 episodes per task — selected by quality score from our production collection. The full corpus is far larger (see Full corpus below) and is available on demand.

Contact: support@zen-o.xyz — for the full dataset, additional tasks, custom object sets, or targeted on-demand collection.

---

What's inside this sample

Robot	Franka Emika Panda (7-DoF arm + parallel gripper)
Format	LeRobot v2.1 (per-episode video files)
Episodes	50 (5 tasks × 10)
Cameras	observation.images.front, observation.images.wrist — 640×480, H.264 mp4 (yuv420p)
Control rate	30 fps
Simulator	MuJoCo

Tasks in this sample (expert set)

Task	Episodes
Bricklayer Palletizing	10
Open Shelf Loading	10
Precision Peg Insertion	10
Sort Parts to Tray	10
Test Tube Rack Loading	10

---

Data schema

Each frame provides:

Key	Dtype	Shape	Description
observation.state	float32	(15,)	7 arm joint positions (rad), gripper width (m), end-effector position (xyz, m), end-effector orientation (quaternion, wxyz)
action	float32	(8,)	7 arm joint targets, gripper command (0–1)
observation.images.front	video	(480, 640, 3)	Fixed front-facing workspace camera
observation.images.wrist	video	(480, 640, 3)	Gripper-mounted wrist camera
timestamp	float32	(1,)	Seconds from episode start
frame_index, episode_index, index, task_index	int64	(1,)	LeRobot indexing

observation.state layout (15): [j1..j7, gripper, ee_x, ee_y, ee_z, ee_qw, ee_qx, ee_qy, ee_qz].

---

Preview / Visualize

Open any episode in the LeRobot dataset visualizer (synchronized front + wrist video with the state/action timeseries):

Task	Episode	Open in visualizer
Bricklayer Palletizing	0	https://huggingface.co/spaces/lerobot/visualize_dataset?dataset=zeno-labs%2Fsim-teleop&episode=0
Open Shelf Loading	10	https://huggingface.co/spaces/lerobot/visualize_dataset?dataset=zeno-labs%2Fsim-teleop&episode=10
Precision Peg Insertion	20	https://huggingface.co/spaces/lerobot/visualize_dataset?dataset=zeno-labs%2Fsim-teleop&episode=20
Sort Parts to Tray	30	https://huggingface.co/spaces/lerobot/visualize_dataset?dataset=zeno-labs%2Fsim-teleop&episode=30
Test Tube Rack Loading	40	https://huggingface.co/spaces/lerobot/visualize_dataset?dataset=zeno-labs%2Fsim-teleop&episode=40

Episodes are grouped 10-per-task in the order above (0–9, 10–19, …, 40–49).

Loading

from lerobot.datasets.lerobot_dataset import LeRobotDataset

ds = LeRobotDataset("zeno-labs/sim-teleop")
print(ds.meta.total_episodes, ds.meta.total_frames)
sample = ds[0]
# sample["observation.images.front"], sample["observation.state"], sample["action"], ...

---

Full corpus & on-demand collection

The public sample above is a small slice. Our full collection currently holds ~26,500 quality-scored trajectories across 12 tasks (bot/sybil-flagged accounts excluded), and grows continuously:

Expert tasks (long-horizon / contact-rich)

Task	Available
Bricklayer Palletizing	2,806
Precision Peg Insertion	2,435
Sort Parts to Tray	2,257
Open Shelf Loading	1,873
Test Tube Rack Loading	1,668
Bin Picking: Extract the Red Part	627
Expert subtotal	11,666

Basic tasks (single-skill pick / place / push / stack)

Task	Available
Lift & Hold	2,855
Pick & Place to Bin	2,681
Stack Two	2,502
Sort to Two Zones	2,387
Push to Zone	2,263
Stand the Peg Upright	2,161
Basic subtotal	14,849

Total: ~26,500 trajectories across 12 tasks.

We can also run on-demand collection for new tasks, custom object sets, alternative camera configurations, or specific success criteria, and deliver in LeRobot v2.1 (or v3.0 on request).

Get in touch: support@zen-o.xyz

---

Notes

• Each episode is a complete, scored task attempt. The public sample is filtered to high-quality successful demonstrations.
• Trajectories from accounts flagged as automated/multi-account are excluded from all released data.
• States/actions are recorded from the live simulation; camera video in this release is rendered offline from the recorded simulation state (same physics, deterministic replay).