Perceptron launches Egocentric 1.0, an API turning egocentric video into robot training data
On November 5, Perceptron officially launched Egocentric 1.0: a self-serve API for annotating first-person (egocentric) video, converting egocentric footage from humans and robots into structured robot training data. It arrives as egocentric video output surges and annotation becomes the bottleneck.
Confirmed
- Outputs include: step-by-step subtask timelines (e.g., reach, grasp, lift, hold), independent continuous activity labels per hand/gripper, and SOTA-level 2D/3D hand reconstruction — addressing the detail loss of assigning a single label to an entire video
- Fine-grained subtask annotation quality is benchmarked against Astra at about 1/8 the cost, priced at $29 per hour of video
- This is Perceptron's first self-serve API endpoint
Why It Matters
- Egocentric video is a key data source for robot manipulation learning, but high-quality fine-grained annotation has long been the bottleneck. Ego 1.0 turns annotation into an API and delivers near-frontier annotation quality at a dramatically lower cost (about 1/8), potentially lowering the barrier to acquiring robot training data.
2026-11-05 ~ 2026-11-05 · 7 related posts
Primary sources
- [source] Perceptron launches Ego 1.0, an egocentric video annotation API at ~1/8 the cost — AkshatS07 · 2026-11-05
- [source] Perceptron Egocentric 1.0 turns egocentric video into robot training data at $29/hour — rohanpaul_ai · 2026-11-05
- Perceptron's egocentric video labeling system boosts hand reconstruction accuracy 12-25% on HOT3D and ARCTIC — SucceededMind · 2026-11-05
- Egocentric video annotation: GPT-6 costs $100+/hour, Perceptron claims 8x cheaper — AkshatS07 · 2026-11-05
3 near-duplicate retellings: AkshatS07 · AkshatS07 · AkshatS07