Safety Manifold for Stochastic Model Predictive Control of an Autonomous Vehicle Approaching Unsignalized Crosswalks with Pedestrians

The paper deals with design of a safe speed control strategy for autonomous vehicle approaching a crosswalk with pedestrians, with an emphasis on its extension with a safety manifold. A model predictive control (MPC) designed to minimize deviation from a preferred cruising speed and maximize driving comfort is used as the safe speed control backbone. The uncertainties related to pedestrian crossing decisions are accounted for by employing a stochastic MPC (SMPC) and a stochastic prediction model of pedestrian crossing behavior. The safety manifold is introduced to ensure robustness with respect to prediction model errors. It is realized by including an additional position-dependent upper limit on the vehicle speed into the SMPC formulation. Its main idea is to keep the vehicle speed below this limit until realization of pedestrian decision, to be able to stop and avoid collision if the pedestrian opts for crossing. Functionality of the proposed safety manifold and related safe speed control is demonstrated through large-scale Monte Carlo simulations of single-vehicle/single-pedestrian interactions.