Labelled robot video, back before your next training run.

Send your recordings. Talo returns boxes, tracks, action segments and task descriptions in the schema your pipeline already reads.

Request a pilot See a labelled sample

Before

Before. A raw frame, as you would send it.

After action: pour

After. Object tracks and an action label, from the sample below. right handbottlecup

Problem

Robot experiments stall while recordings wait for labels.

Ten hours of footage is 1,080,000 frames to label.

That is 36,000 seconds at 30 fps. Each frame needs boxes, tracks or an action label before training can start.

Labels come back with no rules and no check record.

So you can't tell which labels to trust, or cite how they were made.

Every export needs converting.

Someone on the team rewrites the vendor's format before the pipeline can read it.

How it works

Labelling moves off your team's desk.

Today

Record
Wait for labelsStall
Convert formatStall
Train

With Talo

Record
Talo labels, checks and formats
Label. Boxes, tracks, action segments and task descriptions.
Check. Checked against the rules you agreed, with a check log.
Format. Delivered in the schema your pipeline already reads.
Train

1
Send a brief
Robot, tasks, the labels you need and a few sample recordings.
2
Agree the rules
We write labelling rules with you: action names, where each segment starts and ends, and what counts as a failure.
3
Check a pilot batch
We label a small batch. You review it against the rules before any larger order.
4
Receive each batch
Labels in your schema, with the rules and a check log.

One episode, as delivered

Pick up the bottle with the right hand and pour water into the cup.

grasp0.0 to 2.5 s
tilt2.5 to 4.5 s
pour4.5 to 6.5 s
return6.5 to 8.0 s

Tracks right hand, bottle, cup
Segments 4, contiguous, no overlap
Check 28 frames, 9 corrections logged

"segments": [
  { "start_s": 0.0, "end_s": 2.5,
    "action": "grasp",
    "description": "Right hand closes on the bottle." },
  ...
],
"review": {
  "frames_checked": 28,
  "threshold_px": 4,
  "corrections": [ ... ]
}

From the sample below. Field names follow your schema; this file uses Talo's default.

Who it's for

Teams that train manipulation policies on video.

Academic robotics labs

Your students run experiments while Talo labels the frames.

You send: Teleoperation or egocentric recordings, and the labels your paper or policy needs.
You get: Tracks, action segments and task descriptions, with the rules written down so the method section can cite them.
Common start: Clearing a backlog, or re-labelling an existing dataset to a new action list.

Humanoid and robotics startups

Labels that keep pace with your robots and your data vendors.

You send: Recordings from your robots or vendors, and the schema your training code reads.
You get: Tracks, action segments and task descriptions, delivered in your training format.
Common start: A pilot batch on two or three tasks.

See the case study Need recordings too? See collection

Sample