Dataset Viewer
The dataset could not be loaded because the splits use different data file formats, which is not supported. Read more about the splits configuration. Click for more details.
Couldn't infer the same data file format for all splits. Got {NamedSplit('validation'): ('json', {}), NamedSplit('test'): ('imagefolder', {})}
Error code:   FileFormatMismatchBetweenSplitsError

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.

Cross-Episodic Sim β€” v4

850 successful simulated household demonstrations, generated in MuJoCo with a Franka arm on a TidyBot base. Each episode uses only the onboard navigation and wrist RGB-D cameras. Five tasks are covered: set_table, clear_and_store, prepare_and_serve_meal, warm_two_pots and unpack_and_sort_groceries. The houses combine RoboCasa kitchens with MolmoSpaces rooms, and all inter-room doors start open.

split episodes houses
fine_tuning (history) 500 (100 / task) frozen 2–5 room houses
test_same_map 200 (40 / task) same houses as history
test_new_map 100 (20 / task) unseen 3–5 room houses
dev_same_map 25 (5 / task) same houses as history
dev_new_map 25 (5 / task) unseen houses, separate from test

Only successful episodes are included. Success is decided by the semantic task scorer.

Layout

<split>/manifest.json          episode list: id, task, folder, objects, duration, start room
<split>/objects.json, layouts/ object assets and per-episode layout images
houses/                        frozen house definitions
memory/                        frozen same-layout memory lists (house, task) -> history episodes
tasks/                         per-task manifests
episode_index.json             episode id -> shard + path inside the shard
all_open_fixed_v4.yaml         generation protocol
episodes/<split>/<task>-NNN.tar.zst   episode folders, ~30 GB raw per shard

Extract shards into one root, and <split>/<folder> from the manifest resolves:

hf download plnguyen2908/cross-episodic-sim --repo-type dataset --local-dir v4
cd v4 && for f in episodes/*/*.tar.zst; do zstd -dc "$f" | tar -xf -; done

Each episode folder contains:

  • composite.mp4: a 2Γ—2 grid with navigation RGB and depth on top, wrist RGB and depth below.
  • head_camera.mp4
  • action_recording/actions.npz: 50 Hz commands and qpos.
  • model.mjb and initial_state.npz: the saved scene.
  • scene.xml
  • skill_calls.json and oracle_plan.json: the skill-level plan.
  • instruction.txt
  • grasp annotations
  • report.json
Downloads last month
28