Real-time State Estimation for Aggressive Driving Autonomous Ground Vehicles

Abstract

State estimation is an integral part of many robotic systems. It is particularly important in the realm of high performance ground vehicles where systems must be real-time, robust, and accurate. This project aims to tackle the problem of improved state estimation reliability, robust to GPS dropout, applied to the control of an aggressively driving autonomous vehicle in an unstructured environment. To accomplish this goal, the system combines components for visual SLAM, wheel odometry, IMU integration, and incremental inference on a factor graph. The result is a system that is tolerant to short periods of GPS dropout allowing a vehicle to safely handle the situation.