Section: New Results

Deep Reinforcement Learning for end-to-end driving

Participants : Raoul de Charette, Maximilian Jaritz, Fawzi Nashashibi.

Following the work initiated in 2017, we continued the work on end-to-end driving using with asynchronous reinforcement learning directly. The network learns to map low level control directly with RGB images. To continue previous works initiated, we have applied recent domain adaptation and evaluated our reinforcement learning (learn in a realistic car game) in open-loop on real video footage, showing promising adaptation results. New outcome also include tests on real data (web footage). This led to a publication in ICRA [25]. This research was partially funded by Valeo.