A*-Guided Scenario-based Stochastic Model Predictive Control for an Autonomous Vehicle in Simulated Pedestrian Environments

IEEE International Conference on Intelligent Transportation Systems (ITSC), pp. 3472–3477

Abstract

An autonomous vehicle must behave reasonably in a complex dynamic environment, especially crowded with pedestrians, in order to reach its goal robustly and safely. Our approach, considering the uncertainties induced by the pedestrian behavior as well as by localizing the vehicle itself, is introduced in this paper. The interacting multiple model Kalman filter (IMM-KF) is applied to detect the most likely behavior of surrounding pedestrians and the generation of their trajectories is executed. The A* algorithm is utilized to find a rough path with sudden changes in path orientation. A set of Dubins curves are used to smooth out the calculated path and a reference trajectory is finally calculated. A scenario-based stochastic model predictive control (SSMPC), considering all of the uncertainties and collision constraints, calculates a reasonable longitudinal acceleration by solving the quadratic programming of the trajectory around the current pose. Furthermore, a pedestrian obstacle reciprocal collision avoidance (PORCA) is used for modelling a simulated, more realistic interaction of the pedestrians to the vehicle. We evaluate the approach in MATLAB for both unstructured and semi-structured environments while considering the real-time capability of the guidance layer within 1 second.

Authors 2

1. Chinnawut Nantabut Aachen Institute of Automatic Control (IRT)

   RWTH Aachen University

   Affiliation as printed

   Institute of Automatic Control (irt), RWTH Aachen University,Aachen,Germany,52074

2. Dirk Abel Aachen Institute of Automatic Control (IRT)

   RWTH Aachen University

   Affiliation as printed

   Institute of Automatic Control (irt), RWTH Aachen University,Aachen,Germany,52074