Real-Time Autonomous Vehicle Automation With 5G-Based Edge Computing and Artificial Intelligence

D. Geethanjali, N. Minh, Prasanna Kumar Lakineni, Veeraiah Maddu, A. Nath, Tamil Thendral MJournal of Machine and Computing 2025 · DOI

TL;DR

This paper proposes a 5G-enabled Edge Computing Model (5G-EECM) for autonomous vehicles (AVs) to overcome the latency issues of traditional cloud computing. By processing time-sensitive tasks like obstacle detection and navigation at the network edge, closer to the vehicle, the 5G-EECM significantly reduces end-to-end delay, leading to improved safety, responsiveness, and efficiency.

What Problem Does This Paper Address?

Autonomous vehicles (AVs) promise to revolutionize transportation by enhancing safety and efficiency. However, a critical challenge lies in real-time decision-making, which is heavily impacted by End-to-End Delay (EED) when relying on traditional Cloud Computing (CC). This delay, introduced by the long distances data must travel to remote servers for processing, can be detrimental for safety-critical tasks like obstacle detection and collision avoidance. While CC offers powerful computational resources, its inherent latency makes it unsuitable for immediate responses. Existing AV systems struggle to meet the low-EED requirements for these critical functions, creating a gap for solutions that can process data faster and control AVs more reliably in dynamic environments.

Methods

The core of this paper is the proposed 5G-enabled Edge Computing Model (5G-EECM). This architecture leverages the low-latency and high-bandwidth capabilities of 5G networks to integrate Edge Computing (EC) with Cloud Computing (CC).

  • Edge Control Unit (ECU): This unit is responsible for processing time-sensitive tasks locally, at the network edge, close to the AV. This includes Obstacle Detection (OD) using Machine Learning (ML), specifically a Convolutional Neural Network (CNN), and Collision Avoidance (CA) using Reinforcement Learning (RL).
  • Cloud Control Unit (CCU): This unit handles less time-critical, computationally intensive tasks like large-scale data analysis and long-term planning.
  • Dynamic Switching: A key innovation is the dynamic switching mechanism between EC and CC. This switching is based on several criteria:
  • Latency Monitoring: The system continuously monitors the total EED. If it exceeds a predefined threshold ($T_r$) for real-time tasks, it switches to EC.
  • Error Conditions: If the edge server fails or degrades in performance, the system switches to the cloud. Conversely, if the cloud fails, the edge takes over.
  • Task-Based Switching: Time-critical tasks are processed at the edge, while less time-sensitive tasks are offloaded to the cloud.
  • Network Conditions: The system monitors network quality (packet loss, delay) and switches to the most reliable option.
  • CNN for OD: A Convolutional Neural Network (CNN) architecture is employed for obstacle detection. This model includes convolutional layers for feature extraction, pooling layers for dimensionality reduction, and fully connected layers for classification.
  • RL for CA: Reinforcement Learning (RL) is used to model collision avoidance. The vehicle acts as an agent learning an optimal policy to navigate safely by maximizing cumulative rewards and avoiding collisions, represented as a Markov Decision Process (MDP).

Datasets

Not discussed in the paper. The experiments were conducted on a custom-built AV on a controlled track, and the CNN model was trained on a dataset of track images and obstacles.

Key Insights and Ideas

  • Edge Computing is Crucial for Real-Time AV Tasks: The paper strongly demonstrates that processing time-sensitive tasks at the edge is paramount for achieving the low latency required for safe and efficient autonomous driving.
  • Dynamic Switching is Key to Balancing Performance: The ability to intelligently switch between edge and cloud processing based on task demands, network conditions, and system health is critical for optimizing performance and resource utilization.
  • 5G as an Enabler: The integration of 5G networks is fundamental to the proposed model, providing the necessary high bandwidth and low latency for seamless communication between the AV, edge servers, and the cloud.
  • Synergy of ML and RL: The combination of ML for perception (OD) and RL for decision-making (CA) provides a robust framework for autonomous vehicle control.

Key Results

The experimental results highlight the significant advantages of the 5G-EECM over traditional CC:

  • Reduced EED: EC consistently showed significantly lower EED, ranging from 8.76 ms to 12.09 ms, compared to CC's 36.87 ms to 53.21 ms.
  • Faster Response Times: EC response times were notably lower (17.22 ms to 24.47 ms) than CC (63.54 ms to 85.21 ms), enabling quicker reactions.
  • Improved Collision Avoidance (CA): EC achieved a success rate of 8-9 obstacles avoided out of 10, a significant improvement over CC's 5-7 success rate.
  • Reduced Track Deviation: EC resulted in consistently lower track deviations (2.33 cm to 4.49 cm) compared to CC (7.41 cm to 9.49 cm), indicating more precise navigation.
  • Lower Energy Consumption: EC consumed less energy (1.83 to 2.42 Wh) compared to CC (3.37 to 4.00 Wh).
  • Better Network Quality: EC exhibited significantly lower packet loss and network delay.
  • Lower Processing Load: EC required a lower processing load (36.04% to 46.47%) compared to CC (65.09% to 75.35%).

Limitations and Future Work

  • Limitations:
    • The study was conducted on a controlled track with a custom-built AV, which may not fully represent real-world complexities.
  • While the paper mentions training data for the CNN, specific details about the dataset and its size are not provided.
  • Future Work:
    • Scaling the model for smart city deployment.
  • Improving AI to enhance decision-making process (DMP) accuracy and reliability in more complex scenarios.

Summary updated

Abstract

Autonomous Vehicles (AV) are revolutionizing transportation, but real-time decision-making remains a challenge due to End-To-End Delay (EED introduced by Cloud Computing (CC) based processing. A 5G-enabled Edge Computing Model (5G-EECM) is proposed to address this problem by processing time-sensitive tasks at the network edge, closer to the AV, reducing EED and improving responsiveness. The architecture uses Machine Learning (ML) for Obstacle Detection (OD) and Reinforcement Learning (RL) for navigation, dynamically switching between Edge Computing (EC) EC and CC based on task demands. The study tested the system using a user-friendly AV on a controlled track, revealing increased response times, reduced average EED, reduced energy consumption, and improved OD accuracy. The results demonstrate that 5G-EECM significantly boosts AV systems' real-time safety and efficiency, making it reliable and scalable for next-generation AV systems.
BibTeX
@article{geethanjali2025real,
  author = {D. Geethanjali and N. Minh and Prasanna Kumar Lakineni and Veeraiah Maddu and A. Nath and Tamil Thendral M},
  title = {{Real-Time Autonomous Vehicle Automation With 5G-Based Edge Computing and Artificial Intelligence}},
  journal = {Journal of Machine and Computing},
  year = {2025},
  doi = {10.53759/7669/jmc202505086},
}
Figure