Add physical-event labels to the robot data you own.
We label robot recordings and validate the result on held-out data from your corpus. You receive a physical-state sequence and a physical-event record.
Send the hardest hour in your dataset.
We'll label it and show you what Haptica can cover. If the result is useful, we plan the full corpus with you.
We reply with a secure upload link. If the data cannot leave your network, say so in the brief. Only send footage you have the right to share.
How the pilot works
Before labeling starts, we agree in writing on the task, dataset, signals, baseline, held-out split, what counts as passing, and the rule for continuing.
- 1
Audit IDs and clocks
- 2
Define the validation plan
- 3
Generate and review labels
- 4
Deliver the labels and integration guide
- 5
Compare against the baseline on the same data and compute
We start with the ID and clock audit, before labeling begins.
Calibrated for your corpus
We calibrate Haptica on your corpus and check it on held-out data before delivery. Each project includes confidence, abstention, and a quality report.
Before the first call
Send what you have. We work out the rest on the call.
- The task, in one sentence
- The dataset: what's recorded, how much, aligned to what
- Instruments on the robot today
- The baseline you'd compare against
- What success means, in a number you already track
We'll come having read it.
Good fit
- Contact-rich manipulation
- Video aligned to actions or proprioception
- Repeated task families
- Enough demonstrations for grouped splits
- Optional reference instrumentation for validation
The best pilots are tasks where physical state changes the next action and is not obvious from video alone. If your task is visually easy for a large policy, our labels may not help. We tried that ourselves, and we'll say so if yours looks like it.