Dataset Preview
Duplicate
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Job manager crashed while running this job (missing heartbeats).
Error code:   JobManagerCrashedError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

action
list
alert
float32
observations
dict
provenance
dict
[0.41662299633026123,0.22563199698925018,0.14395900070667267,-0.010906434617936611,-0.99993848800659(...TRUNCATED)
0
{"images":{"cam_high":[[[127,131,135],[127,131,135],[127,131,135],[128,132,136],[128,132,136],[127,1(...TRUNCATED)
{"action_monotonic_ns":320917870724062,"action_sync_index":3,"action_xarm_causal_fill":0,"action_xar(...TRUNCATED)
[0.4166240096092224,0.22563199698925018,0.14395900070667267,-0.012902287766337395,-0.999914646148681(...TRUNCATED)
0
{"images":{"cam_high":[[[129,130,135],[129,130,135],[130,131,136],[130,131,136],[130,131,136],[129,1(...TRUNCATED)
{"action_monotonic_ns":320917904300453,"action_sync_index":4,"action_xarm_causal_fill":0,"action_xar(...TRUNCATED)
[0.4166249930858612,0.2256309986114502,0.14395900070667267,-0.016913829371333122,-0.9998546838760376(...TRUNCATED)
0
{"images":{"cam_high":[[[129,130,133],[129,130,133],[130,131,134],[130,131,134],[129,130,133],[129,1(...TRUNCATED)
{"action_monotonic_ns":320917937457999,"action_sync_index":5,"action_xarm_causal_fill":0,"action_xar(...TRUNCATED)
[0.4166260063648224,0.2256300002336502,0.14395900070667267,-0.01867954432964325,-0.9998232126235962,(...TRUNCATED)
0
{"images":{"cam_high":[[[129,130,133],[129,130,133],[130,131,134],[130,131,134],[130,131,134],[129,1(...TRUNCATED)
{"action_monotonic_ns":320917970729998,"action_sync_index":6,"action_xarm_causal_fill":0,"action_xar(...TRUNCATED)
[0.4166260063648224,0.2256300002336502,0.14395900070667267,-0.021341001614928246,-0.9997698664665222(...TRUNCATED)
0
{"images":{"cam_high":[[[127,131,135],[127,131,135],[127,131,135],[128,132,136],[128,132,136],[127,1(...TRUNCATED)
{"action_monotonic_ns":320918004133269,"action_sync_index":7,"action_xarm_causal_fill":0,"action_xar(...TRUNCATED)
[0.41662800312042236,0.22562900185585022,0.14395900070667267,-0.02645951509475708,-0.999647259712219(...TRUNCATED)
0
{"images":{"cam_high":[[[127,132,131],[127,132,131],[128,133,132],[128,133,132],[128,133,132],[127,1(...TRUNCATED)
{"action_monotonic_ns":320918037430858,"action_sync_index":8,"action_xarm_causal_fill":0,"action_xar(...TRUNCATED)
[0.41662898659706116,0.22562800347805023,0.14395900070667267,-0.029249422252178192,-0.99956941604614(...TRUNCATED)
0
{"images":{"cam_high":[[[127,131,133],[127,131,133],[128,132,134],[128,132,134],[127,131,133],[127,1(...TRUNCATED)
{"action_monotonic_ns":320918071344791,"action_sync_index":9,"action_xarm_causal_fill":0,"action_xar(...TRUNCATED)
[0.41662898659706116,0.22562800347805023,0.14395900070667267,-0.031245499849319458,-0.99950897693634(...TRUNCATED)
0
{"images":{"cam_high":[[[129,130,135],[129,130,135],[130,131,136],[130,131,136],[130,131,136],[129,1(...TRUNCATED)
{"action_monotonic_ns":320918104056385,"action_sync_index":10,"action_xarm_causal_fill":0,"action_xa(...TRUNCATED)
[0.4166310131549835,0.22562700510025024,0.14395900070667267,-0.03643949329853058,-0.999332845211029,(...TRUNCATED)
0
{"images":{"cam_high":[[[129,130,135],[129,130,135],[129,130,135],[130,131,136],[130,131,136],[129,1(...TRUNCATED)
{"action_monotonic_ns":320918137390912,"action_sync_index":11,"action_xarm_causal_fill":0,"action_xa(...TRUNCATED)
[0.4166319966316223,0.22562600672245026,0.14395900070667267,-0.03923647850751877,-0.9992268681526184(...TRUNCATED)
0
{"images":{"cam_high":[[[127,131,133],[127,131,133],[127,131,133],[128,132,134],[128,132,134],[127,1(...TRUNCATED)
{"action_monotonic_ns":320918170775386,"action_sync_index":12,"action_xarm_causal_fill":0,"action_xa(...TRUNCATED)
End of preview.

RATE Real-World Training Data

Real-world robot training data used by RATE: Risk-Aware Tactile Encoding for Contact-rich Robotic Manipulation.

The public release contains exactly 50 episodes for each task:

Directory Task Episodes
usb_insertion/ USB insertion episode_0.hdf5–episode_49.hdf5
plug_insertion/ Plug insertion episode_0.hdf5–episode_49.hdf5
bottle_cap_screwing/ Bottle-cap screwing episode_0.hdf5–episode_49.hdf5

Each episode uses an EEF10 state/action representation and contains two camera RGB streams, two tactile RGB streams, and a continuous training-only /alert label in [0, 1]. Bottle-cap screwing additionally contains dual tactile-depth supervision. Alert is not an inference input.

The files are stored as HDF5 and are organized by task. Episode numbering is kept unchanged from the verified RATE delivery dataset.

Downloads last month
113

Paper for jiangyuyao14/RATE-real-world-training-data