Observed arrival · 2026-09-19
Bender’s First-Person Robot-Learning Data
Bender commissions egocentric RGB-D recordings of people performing real tasks, then delivers each accepted session as a single MCAP file.
- For
- Robotics and embodied-AI research labs
- Worth noticing
- Each accepted session contains RGB, depth, pose, and IMU channels on a shared timestamp clock.
Field notes
Bender describes a commissioned-data workflow rather than a self-serve dataset catalog. A lab proposes an activity, the service writes a task specification, and contributors record it from their own point of view using phones they already own. The documented output is an MCAP container with color video, depth maps, camera pose, and raw accelerometer and gyroscope data sharing timestamps. The page says accepted sessions are checked for alignment and task compliance before delivery.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue