University of Twente Student Theses


Machine learning for cooperative automated driving

Chandramohan, Aashik (2018) Machine learning for cooperative automated driving.

[img] PDF
Abstract:Most of the research in autonomous driving currently involves using the on-board sensors on the vehicle to collect the information of the surroundings and using that information for controlling the autonomous vehicle. This research investigates how Machine Learning can be used with cooperative driving for self-driving vehicles. Here the self-driving vehicles can use the vehicle information of the surrounding vehicles to manoeuvre around them. Therefore, it does not need to rely on just its on-board sensors to navigate through the traffic. In this research Reinforcement learning is used for designing a driving algo-rithm to control a self-driving vehicle in a highway environment which is simulat-ed using SUMO (Simulation of Urban Mobility). This research focuses on the method to design the driving algorithm which involves choosing the correct input features and actions for the driving agent and the design of the reward structure. It also explains how the performance of the driving algorithm is effected with change in the reinforcement learning parameters. Except the self-driving vehicle, the other vehicles are controlled by SUMO itself. The aim is to check if the self-driving vehicle is able to manoeuvre through the traffic in the highway by over-taking other vehicles and when needed allowing the faster moving vehicles to overtake it. The driving algorithm is trained and tested in a 2 lane highway environment and a three lane highway environment. It was found that having collision detec-tion as just a part of the rewarding policy did not give the desired results, as the collision percentage was found to be in the confidence interval between 20.5% to 28.4% in a 2 lane highway. Hence collision detection and avoidance was done by a separate entity outside the learning algorithm. This helped in reducing the collision percentage to be in the confidence interval of 0.015 and 0.048 for a two lane system and between 0.08 and 0.14 for a 3 lane system. The research also analyses the performance of the driving algorithm in case of packet loss and change in communication range. A method to cope with the packet loss is also discussed in this report. This research can be used as a basis for using cooperative driving for self-driving vehicles, but further research needs to be conducted to make the self-driving vehicles safer and reliable as the collision rates achieved using this method is still significantly high.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:53 electrotechnology, 54 computer science, 55 traffic technology, transport technology
Programme:Embedded Systems MSc (60331)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page