RoboCap is a wearable egocentric video capture device with six cameras and two IMUs that records first-person human activity to train robots and embodied AI.
Photo source:
RoboCap
RoboCap
is a wearable egocentric video capture device for collecting first-person data
on human activity to train robots and embodied AI. Egocentric means recorded
from the wearer's own point of view, which the site describes as "The
world from a human point of view." Embodied AI is artificial intelligence
that acts through a physical body, such as a robot. FrodoBots developed RoboCap
with input from what the site calls leading robotics researchers.
Wearers
record while going about real-world tasks. Six cameras and two IMUs, or
inertial measurement units, record head orientation and movement, and the
company says the device captures full scene context, including eye direction,
hands and the environment. Each camera shoots 1080p video at 30 fps with a
field of view of 150° or more, and the company states latency between all six
cameras is under 1 ms. The two 6-DOF IMUs, which track six degrees of freedom,
sample at 200 Hz.
The
device runs for over 10 hours of continuous use with an external power bank,
and it works with any standard portable power bank. Onboard storage is 215 GB
of eMMC memory, and Wi-Fi handles syncing devices and reaching a control panel.
Outputs are MP4 video from the six cameras, IMU data and per-cap calibration
files.
The
applications named are robotics, AI training and embodied AI development.
RoboCap comes from FrodoBots, and the site lists BitRobot and GSI as partners,
with a GSI API showcase linked. Its shop is organized into collections for
individual deployers and for enterprise research, and a docs site is linked for
setup and support. A data enrichment API is offered as an add-on for enterprise
customers.
Please subscribe to have unlimited access to our innovations.