Paper on Autonomous Driving on Urban Environments via Imitation Learning of Deep Neural Networks
One more paper on Autonomous Driving on Urban Environments, resulting from the work of our Master student Gustavo Claudio Karl Couto, supervised by prof. Eric A. Antonelo, from the AI group at the Department of Automation and Systems Engineering:
Gustavo designed a new architecture (hGAIL) based on Generative Adversarial Imitation Learning for autonomous navigation in a simulated city. In particular, we employed a conditional GAN (Generative Adversarial Network) to learn the mapping between images of frontal vehicle’s cameras and its bird’s-eye view representation (BEV), which is input to the policy network from GAIL. All networks (from GAIL and GAN) learn simultaneously as the agent interacts with the environment.
Our experiments have shown that GAIL exclusively from cameras (without BEV) fails to even learn the task, while hGAIL, after training, was able to autonomously navigate successfully in all intersections of the city.
More on Autonomous Driving here.
