Control of Autonomous Vehicles Using MPC and Machine Vision with Artificial Intelligence
(Room Palms Ballroom - Booth 100)
16 Jul 19
5:30 PM
-
7:00 PM
The poster presents software design integration employed in the autonomous vehicle system. Specifically, the research has been concentrated on the software integration, which includes the design of the software architecture showing how the main functional blocks of the system are integrated with some details of each functional module such as the machine learning, perception, localization and control of the vehicle. The goal is to have the vehicle navigate through straight road driving with traffic sign and traffic light detection, intersection crossing, and waypoint navigation using HD maps, and autonomous parking. These tasks must be executed while abiding by the traffic rules such as traffic signals, signs, and lanes. The longitudinal and lateral motion of the vehicle is controlled by an Adavanced Model Predictive Controller, which was designed and implemented on a car for Lane Keeping Assist (LKA) and Adaptive Cruise Control (ACC) systems, are presented. The goal of the control system is to follow linear trajectories and stay in the lanes by correcting the lateral deviation to reach the destination point. By regulating the longitudinal and lateral accelerations of the vehicle, it is possible to provide hands-free driving experience. Model Predictive Control with the linear time invariant system with input, output and state variables uses the feed-forward and the disturbance models are used for ACC and Lane keeping Assist systems. The assistance system helps in virtually screening of directions using GPS, tracking of sidewalks, identifying the traffic signals and other sign boards. MATLAB/Simulink and TensorFlow are the two main software used in developing this application. TensorFlow is a computational framework used to build machine learning models. Images and other types of data are captured using sensors on Android devices. Then TensorFlow is used to perform deep learning, and, thus identify and classify different traffic warning signs in real-time. Using the Android Studio, the code can be used in the application. Ground robots, Husky by Clearpath and multiple Turtlebots, are being used for validation of the concepts via implementation.