seeed studio USB Accelerator

I got this to experiment with edge AI processing for a robotics project. Specific machine vision and voice command processing. Currently I'm using the Robot Operating System to handle vision (object tracking, facial recognition, etc.) and voice command processing on Linux. While this works it is a bit clunky even on a reasonably powerful desktop. Initial testing with the TPU provides much better processing of voice commands to control my robot head. And separate vision processing tests are also much smoother processing the video stream from the eye camera in the head. I plan on adding another TPU so one will process the voice commands and the other will handle the vision processing. And depending how how that goes I'll likely add another TPU for some additional machine learning related processing. The only "con" I've see so far is that the TPU does get pretty warm running at the default clock speed and very hot when running at full speed. So you want to make sure it is well cooled and not where you can accidentally touch it when running at full clock speed.

Works well out of the box. Increased object detection FPS from 5 to 20-30. I followed edje electronics YouTube video on using it with a rpi 4.

Good did a fantastic job to develop such a tiny chip to do deep learning inference for computer vision applications. I could implement a couple of models with Google Coral Accelerator and found that a useful and reliable device.

Used with a Raspberry Pi 4 and Tensorflow Lite. Works great!