Object-following autonomous car

I've been working a lot with my AI-enabled embedded camera for work and fun. After focusing on real-time object detection, I decided to give it a funny spin and integrate the camera into a toy car to make it drive autonomously.

As you can see in the demo video, the car has a camera on the front that scans the nearby surroundings looking for a penguin peluche: when the penguin is localized, the car drives toward it. If the penguin cannot be detected, the car rotates to the left and to the right to expand its field of view.

Considerations

This is a straight-forward setup where a single microcontroller runs the object detection algorithm and drives the motor at the same time. The detection happens pretty fast (~50ms), so this setup can work fine in a lot of use cases. If extreme speed is required, I later experimented with a dual-boards setup, where the camera only does the detection and communicates with another microcontroller, who drives the motors, the coordinates of the object of interest.

By leveraging higher-end boards (still in the 30$ price range), the detection can happen in as little as 10ms and the driver board can perform any other kind of task without any impact on performance and responsiveness.