Scholarlink

The major problem of urban transportation network is traffic congestion which leads to increased travel times. This problem may be solved when the real time routing information along with the future traffic information is available with the travelers. To address the issue, we present a novel route optimization method. This method integrates Capsule Networks for traffic prediction with Deep Q-Network (DQN)-based Reinforcement Learning (RL) to enhance real-time vehicle routing. We also proposed to use a central global controller to make traffic decisions considering the traffic scenarios of the complete network. The traffic forecasts generated by Capsule Networks inform the DQN-based route guidance system. It dynamically reroutes vehicles to minimize travel times and queue lengths. We have taken the traffic scenario of a local city in India, obtained from Open Street Map and simulated it with the help of a simulator. We also compared the proposed algorithm with two existing algorithms of vehicle routing, the three key evaluation metrics Average Travel Time (ATT), Queue Length (QL), and Vehicle Throughput are used to compare the proposed method against the compared algorithms. The results demonstrate that our method reduces ATT and QL by 15–25%, and significantly improves traffic flow compared to existing approaches. It highlights its potential for real-world implementation in intelligent transportation systems.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hybrid Approach to Dynamic Vehicle Routing: Integrating Capsule Networks and Reinforcement Learning for Traffic Prediction and Optimization

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sourabh Dave,
  • Pragya Shukla

摘要

The major problem of urban transportation network is traffic congestion which leads to increased travel times. This problem may be solved when the real time routing information along with the future traffic information is available with the travelers. To address the issue, we present a novel route optimization method. This method integrates Capsule Networks for traffic prediction with Deep Q-Network (DQN)-based Reinforcement Learning (RL) to enhance real-time vehicle routing. We also proposed to use a central global controller to make traffic decisions considering the traffic scenarios of the complete network. The traffic forecasts generated by Capsule Networks inform the DQN-based route guidance system. It dynamically reroutes vehicles to minimize travel times and queue lengths. We have taken the traffic scenario of a local city in India, obtained from Open Street Map and simulated it with the help of a simulator. We also compared the proposed algorithm with two existing algorithms of vehicle routing, the three key evaluation metrics Average Travel Time (ATT), Queue Length (QL), and Vehicle Throughput are used to compare the proposed method against the compared algorithms. The results demonstrate that our method reduces ATT and QL by 15–25%, and significantly improves traffic flow compared to existing approaches. It highlights its potential for real-world implementation in intelligent transportation systems.