Abstract :
Computer-based guidance of passenger vehicles is a common reality today, but cost, computation, and robustness challenges remain to obtain accurate vehicle state estimates. This study builds on previous work by the authors towards the development of a vehicle state estimation framework that uses optimal preview control theory to fuse map, GPS, inertial, and forward-looking camera information in a linear filter that offers a-priori predictions of state estimate accuracy. By designing an optimal preview controller around a preview filter designed to make full use of a test vehicle´s low-cost sensors, on-board map, and available visibility, a matched perception and control system is obtained. The resulting preview-based guidance system has a structure similar to LQG algorithms, and is tested both in simulation and on a real vehicle. The closed loop system provides lane-level tracking performance with low cost sensors.
Keywords :
Global Positioning System; cameras; control engineering computing; control system synthesis; linear quadratic Gaussian control; object tracking; predictive control; road vehicles; sensors; state estimation; traffic engineering computing; GPS; LQG algorithms; forward-facing monocular camera; inertial information; lane-level tracking performance; linear filter; low cost lane-following system design; map fusion; on-board map; optimal preview control design theory; passenger vehicle computer-based guidance; preview-based guidance system; temporal preview estimation; test vehicle low-cost sensors; vehicle state estimation framework; visibility; Cameras; Equations; Geometry; Mathematical model; Roads; Sensors; Vehicles;