Vehicle Control in GNSS-Denied Environments Using Map-Based Localization and Model Predictive Control

In this paper, we present the combination of a semantic LiDAR localization algorithm and a Model Predictive Control (MPC) to ensure safe driving in GNSS-denied environments. We use two different sources for the localization: a commercial, standardized HD map and OpenStreetMap. Coupling a graph-based optimization and a Monte Carlo Localization results in an accurate position of the vehicle. Based on this position and the vehicle's velocity, the MPC calculates the desired steering angle and the acceleration to follow a predefined path as accurately and smoothly as possible. We conduct real-world experiments using our autonomous testing vehicle and a commercial map of our campus. Additionally, we assess the outcomes of controlling the vehicle using a high-precision INS and compare it to our localization approach. With a mean deviation to the reference path of 0.069 m, we prove that the combination of a GNSS-free, global, LiDAR localization and the MPC is a suitable setup for autonomous vehicles in GNSS-denied environments.