Data-Driven Hardware-in-the-Loop Plant Modeling for Self-Driving Vehicles

Abstract

In this paper, we present data-driven hardware-in-the-loop (HIL) plant models of different subsystems of a self-driving vehicle. Despite numerous concerns, the automotive industry is still investing remarkable resources into the production of self-driving vehicles. Among the challenges in the development process of such vehicles are the validation and testing process of various subsystems. Here we provide data-driven models of different subsystems so that the automotive industry can validate and test autonomous vehicles without the need of a physical vehicle, which would reduce the considerable amount of cost to the automotive industry. The vehicle subsystems considered in this work include the steering, acceleration, brake, shift, speed, and speed control subsystems. Each of these subsystems is either a multi-input single output or single-input single output system. A Lexus RX450H self-driving vehicle is employed to collect raw data (inputs and outputs data for different subsystems) offline. We used the deep learning toolbox available in the commercial software package, MATLAB/SIMULINK, for modeling each of these systems. The contribution of this paper is twofold. First, collecting real time raw data from a physical Lexus RX450H vehicle and using it to develop machine learning models to represent the vehicle subsystems. Second, subsystem models created using machine learning tools for the Lexus vehicle are tested using Hardware-in-the-Loop. Therefore, the results of such modeling could be used for validation and testing without the need for a physical self-driving vehicle. The proposed modeling results could be useful for reducing the cost of the vehicle development process, since a physical vehicle is not required for validation and testing.