Demonstrations designed around robotic manipulation.
Explore Universal Manipulation Interface (UMI)-style data collection for dexterous tasks, with gripper-view footage and the motion signals required by the target learning workflow.
Bridge everyday demonstration and robot action.
UMI-style collection uses instrumented, hand-operated gripper setups to capture manipulation demonstrations without running every collection on the target robot. The exact hardware, pose representation and signals are project dependent.
View from the tool
Gripper-mounted camera footage focused on interaction between gripper and object.
Manipulation trajectories
End-effector pose, opening width and temporal alignment where capture hardware and processing support them.
Task-specific protocols
Define actions, objects, success states, repetitions and reset conditions before capture.
How we scope and deliver.
Your target task, available source signals and acceptance criteria determine what is included in a dataset.
Capture setup
Handheld gripper hardware, camera placement, tracking and calibration defined for each collection.
Possible signals
RGB views, time, gripper width, pose trajectories and episode metadata, subject to verification.
Task families
Pick-and-place, sorting, assembly, tool-adjacent manipulation, bimanual sequences and similar skills.
Delivery
Structured demonstrations and documentation mapped to an agreed training format.
UMI is a specific manipulation-data approach, not a synonym for standard EGO video. Pose trajectories, gripper width and calibrated streams must come from a compatible capture pipeline.
Common questions.
Is every EGO video a UMI dataset?
No. UMI-style data uses a particular gripper and tracking approach; ordinary head-mounted or handheld video does not automatically contain robot-oriented manipulation signals.
Can you provide dual gripper data?
Bimanual collection can be evaluated where suitable hardware, task and supplier capacity are confirmed.
Can we specify our target robot?
Yes. Share the target embodiment, observation space and action requirements so feasibility can be evaluated.
Start with your training objective.
Tell us the data modality, hardware or camera setup, target task, volume and deadlines. We will scope a realistic pilot.