END-TO-END DRIVE BY-WIRE PID LATERAL CONTROL OF AN AUTONOMOUS VEHICLE USING CONVOLUTIONAL NEURAL NETWORKS

Abstract

Autonomous Vehicles has been a topic which has been highly researched in recent times. The number of vehicles with Autonomous capabilities, specifically Advanced Driver Assistance Systems (ADAS) has been increasing in the last few years. All passenger vehicles which come out in the market today are equipped with the basic ADAS features. These features help reduce the work done by the driver and at the same time help in reducing the risk of accidents on the roads. Though a lot of focus has been given on these vehicles, not much attention has been given to vehicles which were released a few years ago without these features. Another major issue which has been plaguing the development of Autonomous vehicles is seamless switching between manual and autonomous driving modes at variable speeds. The vehicles which are manufactured at present require the driver to travel at a constant speed on cruise control (typically at speeds above 30 miles per hour), only after which ADAS features such as steering control and lane assist can kick in. In this work, we propose a method to implement ADAS features in vehicles which lack such features. We also try to simultaneously solve the problem of seamless switching between human and autonomous driving modes, specifically autonomous steering control at variable speeds. In this work, steering control of vehicles using voltage spoofing (can be extended to the throttle and braking modules), development of PID controller for the subsystems, and implementation of End-to-End driving to enable autonomous driving at variable speeds have been discussed. The controller parameters have been fixed by searching the stabilizing set and by implementing transfer learning, the aforementioned problems have been tackled.