A Reinforcement Learning approach for pedestrian collision avoidance and trajectory tracking in autonomous driving systems

Pedestrian collision avoidance is a relevant safety aspect for autonomous driving systems operating in urban scenarios. This paper presents a Reinforcement Learning approach to endow the resulting agent with the following two competing capabilities: managing unexpected pedestrian crossings and tracking a specific trajectory. In particular, we use the Deep Deterministic Policy Gradient, a model-free off-policy algorithm for learning continuous actions. The effectiveness of the proposed Reinforcement Learning system and the associated training approach is demonstrated by means of numerical simulations.