Home · Open data · Manipulation · Robot video
ManipArena training dataset
About 188 hours and 10,812 real bimanual robot trajectories across 20 tabletop and mobile tasks from the CVPR 2026 ManipArena benchmark; Apache 2.0.
Published by ManipArena (Sun Yat-sen University, X Square Robot and partners), 2026-03
Apache 2.0
The publisher's licence permits commercial use and redistribution, with attribution. Hub card tag apache-2.0 and a licence section reading Apache License 2.0. The gate collects contact details and approves automatically.
The host asks you to sign in to Hugging Face and accept the publisher's terms before it serves files.
hf download ManipArena/maniparena-dataset --repo-type dataset --local-dir ./maniparena-datasetGated 'auto': accept the conditions once while logged in, then a token works. 253 GB including the three simulation task folders. Hours are the paper's figure for the real trajectories.
Specification, as published
- Hours (stated)
- 188 h
- Episodes (stated)
- 10,812 episodes
- Scale
- 20 tasks, 10,812 expert trajectories, 13.5M frames, and approximately 188 robot hours
- Task
- Pick & place, Sorting & packing, Tidying & put-away, Assembly, Laundry & folding, Other
- Environment
- Indoor – lab
- Modality
- RGB, Proprioception, Actions, Language
- Capture
- Exocentric
- In frame
- Robot hands
- Captured in
- Undisclosed
- Embodiment
- fixed bimanual tabletop robot and a mobile bimanual robot (models not named)
- Frame rate
- 20 fps
- Resolution
- 640x480
- Formats
- LeRobot v2.1, Parquet, MP4
- Size
- 253 GB
- Hosted on
- Hugging Face
Figures are the publisher's own. Kinetic Blocks has not measured this dataset.
About this dataset
Training data for ManipArena, a real-world benchmark for reasoning-oriented generalist manipulation at the CVPR 2026 Embodied AI Workshop. Twenty real-robot tasks in three groups: tabletop execution reasoning (inverted cup triangle, spoon to bowl, ring onto rod, items into drawer, pouring water, inserting a wire), tabletop semantic reasoning (sorting headphones, classifying by shape, pressing buttons in order, pairing items) and mobile manipulation (clothes into a hamper, hanging a picture, organising shoes, setting tableware). Each trajectory has front and two wrist cameras at 640x480 and 20 Hz with joint, end-effector and base states. Three further task folders (about 1,662 episodes) are simulation and are not counted in the hours.
Publisher's own task names: arrange cup inverted triangle, put spoon to bowl, put ring onto rod, put items into drawer, pour water from bottle, insert wireline, sort headphone, classify items as shape, press button in order, pair up items, put clothes in hamper, hang up picture.
What Kinetic Blocks did, and did not do
Indexed and linked from the publisher. Not hosted, verified or graded by Kinetic Blocks. The publisher's terms govern. The licence shown is the one stated on the publisher's page, linked above. Nothing on this page is a claim by Kinetic Blocks.
Want the verified ones too?
Request access to the marketplace, where verified listings are checked against their files and graded A+ to F, and your money waits until the data holds up.