chocopan-t3-reverse-oracle-rlds-v3

Synthetic scripted-oracle demonstrations of reverse manipulation tasks in simulation -- take an object out of a container or off a plate and put it back on the table -- in the RLDS / TFDS layout that OpenVLA-OFT uses for LIBERO data. This is the largest of the reverse datasets here and the one to use; v1 is an earlier, smaller subset of the same idea.

Contents

Format	RLDS / TFDS, 1.0.0/, 512 tfrecord shards
Episodes	10,611 (3,537 demonstrations x 3 instruction paraphrases)
Transitions	1,440,981
Tasks	45 reverse tasks derived from LIBERO / LIBERO-plus base tasks
Language	355 distinct instructions; each demonstration appears with 3 paraphrases
Images	256x256 JPEG, third-person + wrist
Size	about 53.4 GB

Demonstrations by rendering domain:

Domain	Demonstrations
canonical (unperturbed LIBERO scenes)	1,573
camera viewpoint perturbation	673
robot initial-state perturbation	646
lighting perturbation	332
table-texture perturbation	313

How it was generated

A scripted oracle drives LIBERO's ControlEnv with the OSC_POSE controller at a 20 Hz control frequency: it plans a grasp on the target object, lifts it out of its container or off its support, and places it in a goal region on the table. The task definitions are BDDL files derived from LIBERO / LIBERO-plus forward tasks by swapping the initial and goal predicates.

An episode is accepted only if the goal predicate is satisfied and no non-target object moved by more than 1 mm.

Rejected attempts are dropped during the RLDS conversion, so every episode in this dataset is a success.

The per-episode seeds and the source HDF5 path are recorded in episodes_index.json. The raw HDF5 of the first, canonical batch is published as chocopan/chocopan-t3-reverse-oracle-hdf5-v1; the later batches were not published in raw form and are regenerable from the recorded seeds.

Schema

The layout is byte-identical to openvla/modified_libero_rlds (libero_*_no_noops), which is what OpenVLA-OFT's LIBERO data pipeline expects:

Field	Type
steps/observation/image	(256, 256, 3) uint8, JPEG-encoded -- third-person view
steps/observation/wrist_image	(256, 256, 3) uint8, JPEG-encoded -- wrist view
steps/observation/state	(8,) float32 -- end-effector position (3), axis-angle orientation (3), gripper qpos (2)
steps/observation/joint_state	(7,) float32
steps/action	(7,) float32 -- end-effector delta pose (6) + gripper (1)
steps/language_instruction	string
steps/{is_first, is_last, is_terminal, reward, discount}	RLDS bookkeeping
episode_metadata/file_path	string -- the source HDF5 file

Images are stored in the OpenVLA convention (rotated 180 degrees relative to the simulator's OpenGL output).

Loading

hf download chocopan/chocopan-t3-reverse-oracle-rlds-v3 --repo-type dataset --local-dir ./chocopan-t3-reverse-oracle-rlds-v3

import tensorflow_datasets as tfds   # 4.9.3 -- the version OpenVLA-OFT pins

builder = tfds.builder_from_directory(builder_dir="./chocopan-t3-reverse-oracle-rlds-v3/1.0.0")
dataset = builder.as_dataset(split="train")

for episode in dataset.take(1):
    for step in episode["steps"]:
        image = step["observation"]["image"]          # (256, 256, 3) uint8
        wrist = step["observation"]["wrist_image"]    # (256, 256, 3) uint8
        state = step["observation"]["state"]          # (8,) float32
        action = step["action"]                       # (7,) float32
        text = step["language_instruction"]

To train with OpenVLA-OFT instead, register the builder name recorded in 1.0.0/dataset_info.json in the OpenVLA-OFT dataset registry (configs.py, transforms.py, mixtures.py) and point --data_root_dir at the directory that contains this folder.

Action / proprio statistics

1.0.0/dataset_statistics_<sha256>.json is a statistics cache seeded with the base checkpoint's action and proprio statistics (Sylvest/openvla-7b-oft-finetuned-libero-plus-mixdata, key libero_10), so that fine-tuning keeps the base model's normalisation instead of deriving a new one from this data.

The sha256 in the cache filename is computed from the absolute path of the directory it was built in, so a freshly downloaded copy will not match: re-seed it at the new path before training. If the training log says Computing dataset statistics rather than Loading existing dataset statistics, the cache was not picked up and the action scale will differ from the base model's.

The top-level dataset_statistics_*.json holds the statistics of this data, computed for reference; it is not the file the training pipeline reads.

Limitations

• Synthetic. These are scripted-oracle trajectories, not human teleoperation: the motion style is the planner's, and it is more stereotyped than a human demonstration.
• Failed attempts are excluded, so the data carries no recovery behaviour and no examples of what going wrong looks like.
• Object initial placements are those of the original LIBERO scenes, which barely vary between episodes. chocopan/chocopan-t3-reverse-oracle-rlds-wide-v1 exists to cover that gap.
• The shipped statistics cache is the base checkpoint's, not this data's -- see above.

Related

Repository	Relation
chocopan/chocopan-t3-reverse-oracle-rlds-wide-v1	same generator, widened initial placements
chocopan/chocopan-t3-reverse-oracle-rlds-v1	earlier version, canonical scenes only; contained in this one
chocopan/chocopan-t3-reverse-oracle-hdf5-v1	raw HDF5 of the first batch, including failures

Sources and license

Simulator and scenes	LIBERO (MIT), LIBERO-plus
Statistics seeded from	Sylvest/openvla-7b-oft-finetuned-libero-plus-mixdata (MIT)

Released under the MIT license. Upstream terms still apply to anything derived from LIBERO / LIBERO-plus; the LIBERO-plus source repository carries no license file, while its Hugging Face distribution is published as MIT.