Network-wide traffic simulation with multi-agent imitation learning

Due to the high complexity of traffic system, traffic simulation is an essential and efficient approach for the analysis and evaluation of the traffic system by modelling traffic flow dynamics and vehicle mobility in response to information and control actions, where networkwide traffic simulation could enable traffic engineers to predict the spatio-temporal movement patterns of vehicles and develop network traffic management strategies to alleviate traffic congestion (Mahmassani, 2001, Kim and Mahmassani, 2015). However, building conventional traffic simulation models is often time-consuming due to complex parameter estimation and calibration processes needed for a high-fidelity simulation model. Therefore, data-driven simulation has gained considerable attention in the recent decade with the increasing availability of high-resolution vehicle trajectory data and massive advances in deep learning models. In this paper, we aim to develop a model that leverages both high-resolution trajectory data and deep learning to learn interactions between vehicles and a road network, which can provide the basis for enabling data-driven mesoscopic traffic simulation at the network-level.

Media Info

  • Pagination: 5p
  • Monograph Title: Australasian Transport Research Forum, 8-10 December 2021, Brisbane, Queensland

Subject/Index Terms

Filing Info

  • Accession Number: 01892450
  • Record Type: Publication
  • Source Agency: ARRB Group Limited
  • Files: ITRD, ATRI
  • Created Date: Sep 6 2023 2:05PM