Occlusion-aware On-road Autonomous Driving: A Path Planning Method in Combination with Honking Decision Making

On-road path planning evaluates and tests an autonomous vehicle's perception-reaction particularly when possible hazards occur suddenly on its path. Occlusion-aware path planning is about generating a path that is robust and adaptable to the underlying risks originated from the occluded areas. Most of the existing methods are focused on the cases where typical traffics are occluded. However, suddenly emerged pedestrians on the road, although being educated about how to pass the road, may not be treated in the same way as on-road traffic, because pedestrian motion dynamics, failures to sense approaching vehicles along with their reaction logics are not easy to model. In terms of liability and active safety, a naïve way to handle the complexity of the pedestrians is to consider the worst-case situation, which makes a planner overcautious. Following the predominant first-search-then-optimize framework, this study defines a potential-field-based cost function to measure the underlying appearance of a pedestrian from an occluded region. By requesting the ego vehicle to take the honking operation when it feels risky, we make the honking decisions in combination with the path planning process, thereby avoiding driving overcautiously.