Interaction-Aware Optimal Safe Speed Control for Autonomous Vehicles Approaching Unsignalized Crosswalks With Pedestrians

The proposed autonomous vehicle safe speed strategy is based on a scenario- and grid-based stochastic model predictive control (SMPC) and a probabilistic neural network (NN) model aimed to predict pedestrian behavior when approaching unsignalized crosswalk. The SMPC problem is formulated to minimize the vehicle traveling time, while accounting for vehicle-pedestrian interaction and keeping the risk of collision with pedestrian low. The vehicle control trajectory is conveniently described by only two parameters to be optimized: the vehicle acceleration and the target speed. Apart from reducing the computational complexity, this simplification facilitates the NN prediction model design in terms of lowering the number of model inputs. The proposed SMPC strategy is verified against a baseline control strategy by means of large-scale stochastic simulations. The verification results indicate that the SMPC strategy in average results in significantly lower vehicle traveling time and less aggressive decelerations, while avoiding pedestrian collisions.