Abstract
We propose an efficient method for implementing decision-making functions in automated buses at specific locations. Based on our experience operating an automated shuttle bus, we recognize that the automated bus must make judgments at various points along its route. For instance, it needs to determine whether the traffic signal is green or not, whether passengers are waiting at a bus stop or not, and whether pedestrians will cross the road at a crosswalk or not. Some of these judgments become evident to be necessary after the automated bus begins its operation. To achieve efficient decision-making at specific locations, we propose a method where the automated bus switches simple inference models. These models are trained using thousands of image files captured and stored during the bus's travels near the judgment points. Our approach effectively enhances the driving decision capabilities of an automated bus following a fixed route.
