Low-Cost Autonomous Vehicle using Off-Board Sensors Connected over 5G : Extension of an Autonomous Vehicle’s operational domain design

Sammanfattning: Autonomous vehicles perceive their environment based on several sensors that are onboard the vehicle. These sensors constantly monitor the vehicle’s movement as well as the environment. There is a wide variety of sensors that can be utilized based on the type of data it provides, accuracy and cost. While not all of them are required, some combination of sensors is required to have a functional and reliable autonomous vehicle. For a robust autonomous vehicle, typically, the sensor quality and accuracy need to be high. Having high-quality sensors drives up the procurement costs and computational requirements, which in turn increases the vehicle cost for manufacturers and customers alike. One way to reduce costs is to limit the number of sensors. However, this also limits the vehicle’s sensing capability and range. A vehicle’s sensing capability and range can be improved with the use of off-board sensors, such as an external camera, placed strategically at crucial points on the road, such as in intersections. These off-board sensors can be connected to an autonomous vehicle over the internet using low-latency communication technologies such as 5G. The problem that this work tried to tackle was how to improve the reliability of an autonomous vehicle while limiting the need for many expensive sensors. It aims to show how a camera placed off-board can be used to complement one or more vehicles’ onboard sensors and achieve an extension of the vehicle’s operational design domain, while relaxing constraints on the onboard sensors. This was investigated by building a physical prototype using a 1/5th scaled car with a Lidar and an Inertial Measurement Unit and extending its sensing capability and range with the use of a camera based off-board sensor. The car was robust enough to navigate and make driving decisions. This also meant that the costs of procuring the hardware needed can be reduced. The minimum distance for a lane merging scenario was first derived mathematically and then compared to experimental data. The experimental findings were consistent with the mathematical model within an 11 percent margin of error.