TY  - JOUR
AU  - Oukebdane, Mohammed Anis
PY  - 2026
DA  - 2026/03/21
TI  - A Decentralised Multi-Agent DRL-based Approach for Pedestrian and Vehicle Traffic Signals Controlling Systems Optimisation
JO  - ICCK Transactions on Mobile and Wireless Intelligence
T2  - ICCK Transactions on Mobile and Wireless Intelligence
JF  - ICCK Transactions on Mobile and Wireless Intelligence
VL  - 2
IS  - 1
SP  - 31
EP  - 43
DO  - 10.62762/TMWI.2025.878487
UR  - https://www.icck.org/article/abs/TMWI.2025.878487
KW  - scalable adaptive traffic signal control
KW  - decentralized multi-agent deep reinforcement learning
KW  - pedestrian-vehicle coordination
KW  - traffic signal optimization
KW  - 6G
AB  - Urban traffic congestion is a major issue that negatively affects mobility efficiency, environmental sustainability and road safety. Many recent methods for controlling traffic signals have used methods based on deep reinforcement learning (DRL) and provided positive results. However, it focused primarily on vehicle flow and have not taken into account pedestrian dynamics due to inherent difficulty related to accurately sensing all pedestrians. As a result of these limitations, recent advances in sixth-generation (6G) localisation technology will provide new opportunities to provide precise, low-latency tracking of pedestrians at signalized intersections, allowing for improved control of pedestrian movements in urban areas. The model proposed in this paper named DRL-based pedestrian-vehicle traffic signal management (DRL-PVTSM), provides a solution to this need by providing a decentralized multi-agent DRL approach that jointly optimizes both vehicle and pedestrian movements at each intersection using independent agents each controlled by deep Q-network (DQN). The agents are provided with a pressure-based reward for optimizing vehicle and pedestrian queue densities and have created safety-penalizing rewards based on pressure from pedestrians that are waiting for the lights to change. The DRL-PVTSM framework has been designed in accordance with the principles of scalability, robustness and real-time applicability to large multi-intersection urban traffic networks. This work demonstrates in extensive simulations performed in SUMO software on multiple network traffic topologies of grid and random layouts that the DRL-PVTSM model provides statistically significant improvements in pedestrian waiting time, vehicle travel delay, and decreases in congestion mitigation and intersection-level safety indicators, thus confirming that decentralized DRL with future 6G will provide a viable method for optimizing the joint operation of pedestrian and vehicle traffic signal systems.
SN  - 3069-0692
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Oukebdane2026A,
  author = {Mohammed Anis Oukebdane},
  title = {A Decentralised Multi-Agent DRL-based Approach for Pedestrian and Vehicle Traffic Signals Controlling Systems Optimisation},
  journal = {ICCK Transactions on Mobile and Wireless Intelligence},
  year = {2026},
  volume = {2},
  number = {1},
  pages = {31-43},
  doi = {10.62762/TMWI.2025.878487},
  url = {https://www.icck.org/article/abs/TMWI.2025.878487},
  abstract = {Urban traffic congestion is a major issue that negatively affects mobility efficiency, environmental sustainability and road safety. Many recent methods for controlling traffic signals have used methods based on deep reinforcement learning (DRL) and provided positive results. However, it focused primarily on vehicle flow and have not taken into account pedestrian dynamics due to inherent difficulty related to accurately sensing all pedestrians. As a result of these limitations, recent advances in sixth-generation (6G) localisation technology will provide new opportunities to provide precise, low-latency tracking of pedestrians at signalized intersections, allowing for improved control of pedestrian movements in urban areas. The model proposed in this paper named DRL-based pedestrian-vehicle traffic signal management (DRL-PVTSM), provides a solution to this need by providing a decentralized multi-agent DRL approach that jointly optimizes both vehicle and pedestrian movements at each intersection using independent agents each controlled by deep Q-network (DQN). The agents are provided with a pressure-based reward for optimizing vehicle and pedestrian queue densities and have created safety-penalizing rewards based on pressure from pedestrians that are waiting for the lights to change. The DRL-PVTSM framework has been designed in accordance with the principles of scalability, robustness and real-time applicability to large multi-intersection urban traffic networks. This work demonstrates in extensive simulations performed in SUMO software on multiple network traffic topologies of grid and random layouts that the DRL-PVTSM model provides statistically significant improvements in pedestrian waiting time, vehicle travel delay, and decreases in congestion mitigation and intersection-level safety indicators, thus confirming that decentralized DRL with future 6G will provide a viable method for optimizing the joint operation of pedestrian and vehicle traffic signal systems.},
  keywords = {scalable adaptive traffic signal control, decentralized multi-agent deep reinforcement learning, pedestrian-vehicle coordination, traffic signal optimization, 6G},
  issn = {3069-0692},
  publisher = {Institute of Central Computation and Knowledge}
}