- Stack
stack/real/noise10mm_iter4750_successβ three real-robot Stack successesstack/sim/stack_v3_noise10mmβ 1,000 successful trajectories with 1 cm waypoint noisestack/sim/play_v1β 100 task-free play episodesstack/sim/stack_v3β Stack Cube v3 (1,000 real-motion trajectories)stack/sim/stack_v1_largercubesβ 2,000 v1 demonstrations with 3β7 cm cubesstack/sim/stack_v1β Stack Cube v1 dataset (2,000 demos)stack/sim/stack_v0β 2,000 demonstrations (replaced 2026-09-24)stack/sim/stack_v0_largercubesβ 2,000 demonstrations (added 2026-09-24)stack/sim/stack_v0_pairedβ 4200 trajectories, 200 files, 96.75 GBstack/real/stack_v0β 20 episodes, 20 files, 6.0 GB
- PushT
- Fruit picking
- Dish and utensils
srlwam
Sim-to-real data for a Franka Panda two-cube stacking task: red cube on green cube, with dataset-specific cube sizes. Simulation data generated in Isaac Lab with Mimic; real data captured on the physical cell.
Stack
stack/real/noise10mm_iter4750_success β three real-robot Stack successes
Three selected Cosmos noise10mm step-4750 rollouts: 1,668 recorded steps,
original HDF5/NPZ data, nine videos and trim metadata. Outcomes preserved as
one success and two success_interrupted; this subset is not a success-rate benchmark.
Recordings and documentation Β· Manifest
stack/sim/stack_v3_noise10mm β 1,000 successful trajectories with 1 cm waypoint noise
Stack v3 controller with independent Gaussian XYZ noise (Ο=1 cm per axis), sampled once per waypoint; orientation unchanged. Forty conditions Γ 25 accepted episodes, two 256Γ256 RGB-D cameras at 10 Hz, paired videos and independent audits.
Dataset and documentation Β· Manifest and hashes
stack/sim/play_v1 β 100 task-free play episodes
50 episodes with randomly placed cubes + 50 empty-scene episodes in the Stack v3 environment. Live GPT-6.1 Sol Low control; 20 seconds per episode; two 256Γ256 RGB-D cameras at 10 Hz. No stacking-success objective or filter.
Data and documentation Β· Manifest and hashes Β· Cube-scene video Β· Empty-scene video
stack/sim/stack_v3 β Stack Cube v3 (1,000 real-motion trajectories)
Codex waypoint-program collection with real-calibrated motion speed. Commanded actions have zero rotation (orientation held at reset). 40 appearance conditions Γ 25 episodes, two 256Γ256 RGB-D cameras at 10 Hz. Published 2026-10-11.
Dataset and documentation Β· File hashes Β· Example video
stack/sim/stack_v1_largercubes β 2,000 v1 demonstrations with 3β7 cm cubes
V1 visual randomization: 400 nominal, 1,000 mild and 600 stronger-shadow episodes. Equal cube sizes sampled per shard, two RGB-D cameras, zero rotation commands and size-aware physical audits.
Dataset and documentation Β· File hashes Β· Preview
stack/sim/stack_v1 β Stack Cube v1 dataset (2,000 demos)
Calibrated against the real setup: fixed 30.48 mm cubes, zero rotation commands, two 256Γ256 RGB-D cameras at 10 Hz. 400 nominal, 1,000 mild and 600 stronger-shadow demonstrations; 100 shards with 20 demos each. Completed 2026-09-24, published 2026-09-26.
Dataset and documentation Β· File hashes Β· Train/validation splits Β· 100-demo preview
Legacy preview grids from the previous stack_v0 revision (before the 2026-09-24 replacement). See the current dataset folder for updated spawn-position comparisons.
stack/sim/stack_v0 β 2,000 demonstrations (replaced 2026-09-24)
Real-aligned cube spawning: X 0.48β0.59 m, Y β0.10β0.10 m; cube sizes 30.48 mm Β±8%. All 2,000 final stacks physically verified. Twenty RGB-D HDF5 batches, 100 demonstrations each, including full state and actions. Full recorded trajectories, without first-frame trimming. See README and SHA-256 manifest.
stack/sim/stack_v0_largercubes β 2,000 demonstrations (added 2026-09-24)
Same wider spawn rectangle, cube sides randomized from 3β7 cm per batch, size-aware nonoverlap spacing and placement heights. All 2,000 final stacks physically verified; 3, 5 and 7 cm pilots passed. Twenty RGB-D HDF5 batches, 100 demonstrations each. See README and SHA-256 manifest.
stack/sim/stack_v0_paired β 4200 trajectories, 200 files, 96.75 GB
The same task, restructured for paired comparison instead of scale: 200 base demonstrations, each
replayed 20 more times under domain randomization from the identical initial state and action
sequence (verified bit-identical across all 200). One file per base demo, episode_N.hdf5
(~455 MB), holding the base plus its 20 variants. No depth channel β see the README inside the
folder for exactly what varies between variants and what's held fixed.
stack/real/stack_v0 β 20 episodes, 20 files, 6.0 GB
Real-robot captures on the physical cell, one file per episode, 111-336 frames each. RGB-D from both cameras (raw stream plus a centre crop), joint and end-effector state. The recorded actions carry no rotation.
PushT
Existing release: pusht/sim/v1.
pusht/sim/v3 β PushT v3 (1,000 trajectories)
Scripted physical pushing; native 256Γ256 RGB-D at 10 Hz, 100 Hz physics, Stack v3 camera mounts and DR. Revised acceptance: 30 mm / 20Β° destination, 10Β° robot drift, initial-image alignment diagnostic.
Data and documentation Β· Manifest
Fruit picking
fruit/sim/fruit_v1_noise10mm β 1,000 accepted fruit-picking trajectories
Open the container and pick up the dark plum. Isaac Lab, Stack v3 domain randomization, 1 cm Gaussian XYZ noise once per waypoint, local control with bounded Sol recovery. 865 automatic accepts and 135 explicit user-reviewed accepts, including one flagged contact-quality warning. Two 256Γ256 RGB-D views at 10 Hz.
Data and documentation Β· Manifest and acceptance labels
Dish and utensils
dish/sim/dish_v1_noise10mm β 1,000 accepted trajectories
Take both plates out of the rack and place them flat on the table. Put the spoon on the grey plate and the fork on the pink plate. 40 conditions, 500 episodes per utensil-first order, 1 cm Gaussian XYZ noise once per waypoint, fixed EEF orientation. Local motion planning with bounded Astra recovery; paired 256Γ256 RGB-D at 10 Hz. Final audit acceptance and original simulator labels are distinguished explicitly.
- Downloads last month
- 114

