Deep Learning Based Lateral Control System
Authors:
- 16 December 2020
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Abstract
Autonomous vehicles have seen advancement in technology over a while. Even then, it still lacks a robust system for the lateral control system of vehicles. Autonomous vehicles are considered the future due to the increasing population and limited infrastructure. But modern innovations in the industry along with research in the deep learning domain have shown promising results. We tried to demonstrate a lateral control system trained on a conventional neural network algorithm. An end-to-end deep neural network is discussed in this paper. The trained model of the proposed network is the re-implementation of the existing model AlexNet and PilotNet. The complexity and size of the novel network are much less than existing models. The model has been tested on a virtual environment trained on a small dataset collected from a simulated environment. Despite the small dataset, model attained an accuracy of 97.42% on known tracks and an accuracy of 86.09% on unknown tracks. The result indicates behavioral cloning and reinforcement learning can be used for driving vehicles in an unknown environment.Keywords
This publication has 8 references indexed in Scilit:
- Deep Learning for Autonomous Driving2019
- On the instrumentation and classification of autonomous cars2017
- A Control Strategy of Autonomous Vehicles Based on Deep Reinforcement Learning2016
- Estimation of measurement of distance to the object by analyzing the blur of its image series2016
- DeepDriving: Learning Affordance for Direct Perception in Autonomous Driving2015
- Autonomous reinforcement learning with experience replayNeural Networks, 2013
- Vehicle Detection and Tracking in Car Video Based on Motion ModelIEEE Transactions on Intelligent Transportation Systems, 2011
- Real time detection of lane markers in urban streets2008