Shotwell launches to solve robotics data labeling with self-trained models and human edge-case handling

_akhaliq · x · 2026-08-06

Robotics data labeling startup Shotwell has launched, aiming to solve the hardest problem in robotics: understanding your data. The founder argues that collecting robotics data without understanding it is pointless, but existing options fall short: VLMs can't provide dense, accurate labels with precise subtask boundaries; in-house annotation teams are operationally heavy; and current vendors charge high prices for low quality. Shotwell combines the best of both worlds by training its own annotation models and letting humans handle ugly edge cases, guaranteeing 100% quality at scale.

Original post →

More from Embodied

Embodied channel →