Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Cooperative intersection control for autonomous and connected vehicles using machine learning

: Ravindra Gugale, Rohan
: Liggesmeyer, Peter; Feth, Patrik

Volltext urn:nbn:de:0011-n-5257200 (13 MByte PDF)
MD5 Fingerprint: e24cc9d0dc7a3b30c379622f4a03c236
Erstellt am: 9.1.2019

Kaiserslautern, 2018, XV, 109 S.
Kaiserslautern, TU, Master Thesis, 2018
Master Thesis, Elektronische Publikation
Fraunhofer IESE ()

One of the challenging issues of the 21st century is dealing with the increasing number of vehicles and thereby their congestion problems. The modern day advanced driver assistance systems and promising futuristic technologies such as autonomous vehicles withstand the potential to ensure much more efficient and safe traffic movements by reduced and / or absent human involvement in the dynamic driving tasks. Adaptive traffic light systems using reinforcement learning have been an active area of research for several years. However, with the advent of autonomous vehicles and advancement in the fields of wireless communications, non – signalized intersections or virtual traffic lights can soon be a reality. The aim of this thesis is to evaluate the concept of a system of systems level intersection control for vehicles using reinforcement learning. Each approaching vehicle towards the intersection is an active participant within this system and the system level controller which can be realized as a part of the intersection infrastructure, governs the maneuver of the intersection for each of the participants in order to maximize the traffic flow efficiency. The thesis aims at laying out this intersection strategy using Q – learning algorithm and Q – networks from reinforcement learning. The concept has been evaluated through several iterations of characterizations, experiments and suitable metrics. This thesis also serves as a feasibility study whether an agent can be trained to control intersections using reinforcement learning. One of the main motivations of this project is that several real – world problems are being addressed using machine learning with the recent research and advancements in the field of artificial intelligence. Reinforcement learning is becoming a popular technique to develop solutions to control problems such as intersection management. For majority of the traffic management projects led by the state agencies, the concepts are validated in their early stages in suitable micro – simulation tools before the on – field implementations and thus the micro – simulator used in the scope of this thesis is the SUMO – ‘Simulation of Urban Mobility’ simulator developed by the Deutsches Zentrum für Luft und Raumfahrt.