Dynamic Edge-Server Data Processing for Vehicle Geofencing

Abstract

Geofencing can be used to contain vehicles to a virtually defined area. A type of geofence that does this is the Model Predictive Geofence (MPG), which provides a way to contain vehicles within a geofence by using precomputed safety information from Hamilton-Jacobi reachability analysis. This enables predictive safety checks during operation, where a vehicle can determine whether its current state will lead to leaving the geofence before the geofence boundary is actually crossed. However, the resulting data can require large amounts of memory, making it difficult to store the full MPG on resource-constrained vehicle computers.

This thesis investigates whether predictive geofencing can be used on vehicle computers without storing the complete geofence data onboard. To address this, Dynamic Adaptive Loading of Spatial Datasets (DALOSD) is proposed, where the data of the MPG is divided into smaller segments that are requested from a remote server based on the vehicles current position and predicted future movement. The loading scheme uses a spatial window to determine which segments to load, keep, and discard, and introduces prioritization and speed limitation mechanisms to reduce the risk of missing required data during operation. The results show that DALOSD significantly reduces onboard memory consumption while still providing access to the required MPG data under suitable network conditions.