Probabilistic Collision Risk Evaluation for Moving Vehicles Based on the Interaction-aware Motion Prediction by Considering the Sensing Uncertainty and Blind Area

Abstract

Intelligent driving assistant systems have been studied meticulously for autonomous driving. The systems have the responsibility for driving itself, such as in an autonomous driving system, it should be aware of its’ surroundings including oncoming moving vehicles, and the autonomous vehicle must be able to evaluate collision risk for the ego-vehicle’s planned motion. However, when recognizing surrounding vehicles using a sensor, there are two significant challenges for an autonomous vehicle to recognize the surrounding vehicles and the uncertainty of measured information and misdetection caused by blind spots. Previous studies have considered the measurement uncertainty, and evaluated the collision risk, based on the motion of surrounding vehicles that were predicted in a separate step. However, since the motion of surrounding vehicles can be changed by the evaluated collision risk because of interactions between vehicles, the motion of the surrounding vehicles should be predicted based on the evaluated collision risk. Moreover, for blind spots, the previous study only generated a risk map for planning, by comparing the path of movement of the occluded vehicle, which predicted using the HD-map and ego-vehicle’s path. Since this approach only compared their geometries, it could not consider the speed or uncertainty of the obstructed view of the vehicle. This dissertation overcame the limitations of previous studies by proposing an algorithm for evaluating the collision risk while also predicting the motions due to the interactions of other vehicles according to the ego-motion whilst considering the limitations of recognizing the surrounding vehicles. The proposed algorithm consists of two stages. In the first stage, the vehicle will be occluded by other obstacles is modeled by an HD-map and sensor visibility. In the second stage, the interaction-aware motion is predicted by optimizing the motion of other vehicles based on a mathematical vehicle model that considers measurement uncertainty, the collision probability with the planned motion of the ego-vehicle is estimated in the second stage while optimizing the vehicle motion. The proposed algorithm was evaluated by testing the vehicle in four scenarios; moving to the side lane when another vehicle is approaching from behind, changing into an occupied lane, testing the vehicle whilst obstructed by surrounding vehicles, and blind recognition view of obstruction in a blind spot in an intersection with alleyways. The experimental results show that the proposed method makes the following contributions; the proposed method probabilistically realizes the that there is a vehicle in an obscure, obstructed view, and views it as a moving vehicle detected by the sensor with its existence probability. Moreover, the proposed algorithm simultaneously predicts interaction-aware motion and estimates the collision risk probability between the planned ego-motion and the motion of other vehicles.