We present a control approach for autonomous vehicles based on deep reinforcement learning. A neural network agent is trained to map its estimated state to acceleration and steering commands given the objective of reaching a specific target state while considering detected obstacles. Learning is performed using state-of-the-art proximal policy optimization in combination with a simulated environment. Training from scratch takes five to nine hours. The resulting agent is evaluated within simulation and subsequently applied to control a fullsize research vehicle. For this, the autonomous exploration of a parking lot is considered, including turning maneuvers and obstacle avoidance. Altogether, this work is among the first examples to successfully apply deep reinforcement learning to a real vehicle.

No abstract

No abstract

No abstract

Autonomous driving is no longer a subject of science fiction. Instead it has become a field of highly topical developments and has already reached numerous milestones. The Audi Autonomous Driving Cup provides a stage for students to participate in this development process. This competition, carried out in Germany, Austria and Switzerland, provides miniature vehicles of normed hardware to the participants which have the task of implementing algorithms for autonomous handling of problems such as lane tracking, obstacle detection, adaptive cruise control, overtaking, turning as well as entering and exiting parking spaces. To present the concept of optimal control based on a single track model in combination with image processing methods we focus in this paper on the parking manoeuvre only. Thereby we show that optimization and control techniques can play a key role in the design of autonomous vehicles.