Jerk limited planner for real-time applications requiring variable velocity bounds

The generation of velocity references for autonomous vehicles is a known complex problem. The degree of complexity depends on the requirements to be fulfilled. First of all, in some applications of interest, trajectories need to be planned in real time. Furthermore, for smoothness reasons one may desire them to be jerk limited and, finally, they must satisfy velocity bounds that are variable along the path. Since planners proposed in the literature do not simultaneously fulfill the above mentioned characteristics, this paper proposes a novel generator which is able to efficiently evaluate almost minimum-time references. Despite the heuristic nature of the planner, an extensive test campaign has proved that obtained solutions are only slightly suboptimal w.r.t. the ones provided by a nonlinear solver, i.e., an algorithm for the global optimization whose computational times are incompatible with the online requirement.