A conceptual demonstration of self-learning eco-speed control at signalised intersection

Achieving fuel-efficient driving is a challenging task in the urban environment where traffic flows are frequently and periodically disrupted by traffic controls. Conflicts between vehicles as a result of lane changing or overtaking increase driving complexity. Optimal driving may be a complex trajectory under such circumstances which is likely unforeseeable by drivers. Although eco-driving is receiving recent attention as a promising solution for reducing the adverse environmental impacts of traffic, it needs considerable improvements in the underlying concepts and technical approaches to strengthen its applicability to wider driving environments. This study presents a novel eco-speed control algorithm to assist fuel-efficient driving at signalised intersections. The proposed algorithm employs Q-learning, a self-learning intelligent agent to optimise the driving speed to minimise the resulting fuel consumption. Q-learning is a model-free machine learning technique that is widely used for searching for an optimal action-selection policy for the finite Markov decision process. A significant benefit of the self-learning approach is that the control system can directly learn from past experience under various environmental conditions. The proposed algorithm is designed to derive the fuel-optimal driving trajectory for an individual vehicle by eliminating idling in the vicinity of an isolated signalised intersection. The performance of the proposed algorithm is evaluated and, compared to conventional driving patterns without the ecospeed control using AIMSUN microscopic simulation. The eco-speed control algorithm demonstrates that the fuel consumption can be reduced up to 18% and a significant reduction in the vehicle idling time as well.


  • English

Media Info

  • Pagination: 13p

Subject/Index Terms

Filing Info

  • Accession Number: 01676251
  • Record Type: Publication
  • Source Agency: ARRB
  • Files: ITRD, ATRI
  • Created Date: Jul 26 2018 10:46AM