A path planning of the ADV is performed. A set of obstacle boundaries are generated for one or more obstacles, where each of the set of obstacle boundaries has a corresponding buffer distance ranging from a predetermined minimum buffer distance to a predetermined maximum buffer distance. A set of paths of the ADV is generated using quadratic programming based on the set of obstacle boundaries in parallel, where each path of the set of paths corresponds to one of the set of obstacle boundaries. A path is selected from successful paths of the set of paths based on a corresponding obstacle boundary having a smallest corresponding buffer distance, where the ADV is at least a predetermined distance away from the one or more obstacles in the successful paths. The ADV is controlled to drive autonomously according the selected path to avoid the one or more obstacles.