Task-specific demonstrations.
Define the actions, environments, and variation that matter. Build a collection brief with instructions contributors can follow.
The scope starts with your research question.
For robotics & physical AI teams
Start with people.
Build toward physical AI.
Task-specific human demonstrations, captured on iPhone and annotated with context.
Circuit board. Scroll down to reveal the 3D mesh; scroll up to return to the human view. Illustrative AI-generated footage.
Useful data starts
with a useful question.
Phone recordings aren’t robot trajectories. Their value depends on the task and learning setup. A pilot is a chance to measure that fit.
Built around your task
More than a recording. The context to work with it.
Shape the collection, annotations, and review around a concrete experiment.
Structured annotations.
Action segments, key moments, and task metadata keep the recording connected to what happened.
Spatial capture.
Every recording includes synchronized RGB, camera pose, and intrinsics. LiDAR-equipped iPhones add depth, confidence, 3D point samples, and depth-grounded hand estimates.
A focused first experiment
Build the question.
We build the dataset.
Small pilots. Specific learning goals. An agreed way to evaluate the result.
- 01
Define the question.
Share your task and learning setup. Agree on the demonstration requirements and what a useful first experiment would show.
A shared definition of useful.
Start with the action, setting, variation, and evidence your team needs.
Agree on scope before collection
- 02
Collect with context.
Build task instructions for phone-recorded demonstrations. Specify the action segments, moments, and metadata to capture alongside them.
The action and its context.
Keep recordings, annotations, and task requirements connected throughout review.
Task-specific annotation requirements
- 03
Review and evaluate.
Review against the agreed brief, select demonstrations, and establish the evaluation and delivery requirements together.
Measure the fit.
Use the pilot to learn whether these demonstrations are useful for your task and training setup.
Human review before dataset selection
A few things
to know.
We’re building the pilot with care. Here’s where things stand.
Ask us something elseStart a conversation
What are you
trying to teach?
Tell us about your team, your task, and what a useful first experiment would look like.
Prefer email? hello@opnflr.comDiscuss a pilot
Start with the problem. We’ll work from there.