Using machine learning and especially reinforcement learning to develop agents that get smarter in time
Researching the autonomous vehicles of the future, from flying robots to self-driving cars
Harvesting the power of deep neural networks so the computer can see the world the way we see it
Object recognition, 3D scene modeling and everything that helps a robot perceive the environment
Teaching machines to
see the world the way we do
We're not describing each object to the computer. No more detecting lines, circles and textures. We just show our agent lots and lots of pictures and let it learn on its own. We do this using deep learning techniques, building neural networks with multiple convolutional layers, that can then be queried from portable or embedded systems.
Self-driving car agent
In this case, we replicated Nvidia's "end-to-end" model, to train it on a driving dataset, in order to simulate steering a car in different road conditions. Using just a camera feed, without any other road sensors or image processing.