Systematically comparing control approaches in the presence of actuator errors

A quantitative comparison of different approaches to control the trajectory of an autonomous vehicle can be achieved by measuring the overall performance of the whole system in closed loop. This work compares two trajectory generation and control approaches for autonomous driving in simulation. One approach is based on a low frequency check of geometric safety distances and a high frequency check of control deviations. The other approach minimizes the probability of colliding with obstacles by explicitly considering actuator uncertainties and solving a chance constraint optimization problem. The comparison is based on systematically applying actuator inaccuracies and searching for event combinations leading to collisions. We demonstrate that the control approach incorporating uncertainties performs better in one class of critical scenarios although its error model is different to the applied inaccuracies.