Stochastic Model Predictive Obstacle Avoidance with Velocity Reduction in Accordance with Chance Constraints in Crowded Environments

Abstract

For obstacle avoidance against randomly moving traffic participants, stochastic model predictive control is promising. In crowded environments, however, feasible trajectories satisfying chance constraints do not necessarily exist; crowding induces a relaxation of constraints that causes deterioration of safety. To address this issue, we developed a velocity control method that decelerates the ego vehicle to a speed that satisfies the chance constraints on the prediction horizon. We conducted numerical simulations of obstacle avoidance and experiments of moving through a crowd comprising vehicles and pedestrians to evaluate the performance. The results indicate that the designed controller can generate a trajectory that mitigates the relaxation of constraints and adapts to various traffic participants.