Sites Inria

Version française

Rencontres Inria Industrie

Project teamt Rits (Inria Paris) -

End-to-end driving with Deep Reinforcement Learning

© INRIA Sophie Auvin - R comme Robot

Although traditional methods require an exhaustive understanding of the scene to drive autonomously, we demonstrate an algorithm using Deep Reinforcement Learning that efficiently learned to drive directly from the raw image pixels without any driving examples. This demonstrates a brand new autonomous driving paradigm.

Keywords: Autonomous driving Deep reinforcement learning