Creating a self-driving car with on-the-edge ML
Project WOMBAT (Working On Machine Based Autonomous Transport) is a state of the art self-driving platform in continuous development. I extensively modified a kid’s ride-on toy Mini Cooper with custom motor circuitry, steering feedback, remote controls, and 3D printed parts. Two camera systems on the front handle depth mapping and localization. A hardware machine learning accelerator enables real-time object detection and avoidance without any internet connection and leverages an image segmentation model I trained with TensorFlow.
Check out the car.
Watch the reveal.
A timeless commercial reinvigorated with deep learning
In 2003, Apple launched its iconic iPod dancing silhouette ad campaign. In 2020, anyone can create their own silhouette videos with the power of on-device machine learning. Silhouette Camera takes advantage of the people segmentation feature in ARKit to cut out the people from the camera and place them in front of a solid background on device and in real time. My solution utilizes the accuracy, efficiency, and speed of ARKit’s segmentation model to create a smooth and precise result.
Photo credit: Apple