King's College London

Research portal

Building a Lane Merge Coordination for ConnectedVehicles Using Deep Reinforcement Learning

Research output: Contribution to journalArticlepeer-review

Omar Nassef, Luis Sequeira Villarreal, Elias Salam, Toktam Mahmoodi

Original languageEnglish
JournalIEEE Internet of Things Journal
Accepted/In pressAug 2020


King's Authors


This paper presents a data-driven framework fortrajectory recommendation in automated and cooperative driv-ing. The considered cooperative driving manoeuvre is lane-mergecoordination, and while the trajectory recommendation can onlybe communicated to the connected vehicles, in computation ofthose recommendations both connected and unconnected vehiclesare taken into account. The data-driven framework is imple-mented centrally, comprising of two main components of aTrafficOrchestratorandData Fusion. TheTraffic Orchestratorpredictsthe safest trajectories for connected vehicles involved in the lane-merge manoeuvre. TheData Fusionincorporates camera detectedvehicles in order to map all vehicles including connected andunconnected. To this end, the recommendations are built usingvarious state-of-the-art machine learning techniques includingdeep reinforcement learning and dueling deep Q-network. Ourevaluations are conducted using the real-system deployed in thetest track, with a mix of connected and unconnected vehicles.The results demonstrate precision of predicted trajectories, andpercentage of successful lane merge achieved deploying differentmachine learning techniques.

Download statistics

No data available

View graph of relations

© 2020 King's College London | Strand | London WC2R 2LS | England | United Kingdom | Tel +44 (0)20 7836 5454