In a new development, researchers at QUT robotics in collaboration with Ford Motor Company have discovered a way to instruct autonomous vehicles to select the camera for navigation.
The research is a finding from a project that examines how cameras and LIDA sensors, generally used in autonomous vehicles, can better understand things around them.
“The main point is to learn which cameras to use at different locations, based on past experience at that location,” stated the senior author of the study. For example, the system might discover that a particular camera works well for tracking the location of the vehicle on a particular section of road, and choose to switch to that camera on succeeding visits to that stretch of the road.
Technically, autonomous vehicles carry heavy dependence on their location, using a range of sensors and cameras.
The understanding of location helps to use map information, which is useful to discover other dynamic thing in the surrounding. For example, a particular intersection might have people passing in a certain way.
Importantly, the understanding of location can be used as a prior information for neural nets carrying out object detection, and thus accurate localization is crucial and this research enables to focus on the best camera at any given time.
Meanwhile, to make progress, the team also needed to devise new ways of finding the performance of an autonomous vehicle navigation system.
The focus is not on how the system performs when it’s functioning normal, but how it works in the worst-case scenario.
The research was undertaken as a part of the larger fundamental research project with Ford to examine how cameras and LIDAR sensors, which are integral to autonomous vehicles, can better learn their surroundings.