在遮挡场景中提升弱势道路使用者安全：智能基础设施与自动驾驶车辆的协同方法

The impact of Automated Vehicles (AVs) on road traffic safety has become the focus of discussions among governmental organizations, academia, stakeholders, and OEMs. Questions about how safe the automated driving features should be and how the road infrastructure should be improved for the arrival of this new technology must be clarified to enable full acceptance by the customers and society and prepare the mobility of future cities. The fundamental architecture of automated vehicles comprises perception, planning, decision, and actuation. The operation of the perception system, which is responsible for understanding the environment in which the vehicle is inserted, relies mainly on the onboard sensors. However, the available ranging and vision sensors, e.g., LiDAR, radar, and camera, have several limitations. Scenarios with occlusion present a real challenge for state-of-the-art perception systems. The occlusion, caused by obstructing the sensors’ detection field, limits the vehicle’s perception ability and inhibits the detection of other road users in the surroundings, especially Vulnerable Road Users (VRUs). Infrastructure composed of Roadside Units (RSUs) equipped with infrastructure-based sensors can overcome the perception limitations of a system based solely on onboard sensors by monitoring the road environment with a larger field of view and reduced sensitivity to occlusion. This paper presents a collaborative approach for smart infrastructures and automated vehicles for vulnerable road users’ collision avoidance. The proposed extended perception system comprises four main modules: traffic monitoring, long-term motion prediction, collision risk assessment, and trajectory planning. In the event of a safety-critical scenario, the infrastructure generates a safe and comfortable evasive maneuver to avoid a possible collision. Hence, the proposed approach provides a complete solution to overcome scenarios with occluded VRUs. It allows AVs to react to a critical situation with a longer time-to-collision than other systems relying only on onboard sensors, increasing the chance of successful avoidance even when implementing smoother maneuvers. This contributes considerably to the safe and comfortable operation of automated vehicles.