Abstract
With the release of open source datasets such as nuPlan and Argoverse, theresearch around learning-based planners has spread a lot in the last years.Existing systems have shown excellent capabilities in imitating the humandriver behaviour, but they struggle to guarantee safe closed-loop driving.Conversely, optimization-based planners offer greater security in short-termplanning scenarios. To confront this challenge, in this paper we propose anovel hybrid motion planner that integrates both learning-based andoptimization-based techniques. Initially, a multilayer perceptron (MLP)generates a human-like trajectory, which is then refined by anoptimization-based component. This component not only minimizes tracking errorsbut also computes a trajectory that is both kinematically feasible andcollision-free with obstacles and road boundaries. Our model effectivelybalances safety and human-likeness, mitigating the trade-off inherent in theseobjectives. We validate our approach through simulation experiments and furtherdemonstrate its efficacy by deploying it in real-world self-driving vehicles.