NewShotwell-Bench: an annotation benchmark that accepts valid alternatives. Read more →
ShotwellShotwell

Annotations for robotics

The best understanding of your robotics data. Labels, annotations, and quality checks for robotics labs, deployment companies, and data vendors.

Every step of every episode, named and checked

A real fold from our pipeline. The letters spell the subtask at each moment. Red marks a step that failed its rubric. Click a step to jump to it.

Episode 0431 · fold_clothing—
Subtask—
Time—
Rubric—
Verdict—
Trusted by
Weave RoboticsUltra Frontier MachinesParametric Trace
And more
“Shotwell annotations improved our training data visibility to allow every QA reviewer to review 6× more operators.”

Labels you can train on and query

Subtask segmentation

Sub-second boundaries for every step, checked against your SOP.

Dense captions

A plain description of what happens in each step, in the words your models use.

Failed pathsOut of distribution

Failure detection and quality checks

Rubric verdicts for every step. Find out-of-distribution data to understand what goes into your training.

All forms of data

Egocentric, teleoperation, UMI, or deployment footage. Send us MP4, MCAP, or LeRobot, and get labels back as JSON or CSV within 24 hours.

Let's talk about your data: team@shotwell.ai

Shotwell