Kerve

Huawei Road Reader

Huawei came to Kerve looking for a way to really show off the AI capabilities of the neural processor in their new Mate 10 phone. We suggested some high tech image recognition using the TensorFlow machine learning framework, and then took this to a whole new level by using that image recognition to drive and navigate a custom-built self-driving car.

In just 5 weeks, we hooked the Mate 10 up to a uniquely adapted remote-controlled Porsche, added in a high quality live camera feed, and built an intuitive interface for the machine-learning software that would drive the car, before showing it off at the Mobile World Congress in Barcelona, achieving global recognition across many of the biggest tech editorials.

Ok, my personal involvement was fairly minimal - my knowledge of robotically adapting cars and machine learning is fairly limited, but I did build a remote monitoring control panel that allowed the team to alway see the live status of the phone, what it was thinking, what it was detecting, and what it was telling the car to do, which I’d say was fairly essential.

Check out the video from Engadget below to see everything in action!