QuaMing releases a volume-ready glasses design for embodied-AI data capture
These glasses record what the wearer sees. The design can be built for volume.
The published design uses two chips that are already in production: Bestechnic BES2800 and Fullhan MC6350. BES2800 runs the system, audio, Bluetooth, Wi-Fi, and power. MC6350 does vision capture, dual ISP, on-device AI, and hardware sync of the sensors.
For robot training, the glasses are specified as a stereo RGB-D capture tool: two 2 MP wide modules, distortion under 2 percent at the edge, factory calibration, and a depth map of 320×240 at 30 fps from the 1 TOPS INT8 NPU on MC6350. IMU sync is specified at under 1 ms. The article puts the unit cost at about one fifth to one tenth of a lab capture rig.
The same two-chip base is also described for consumer glasses and a small AI pin. Consumer glasses on this site use MC6350.
