Periodic Reporting for period 1 - ASCENT (Autonomous Vehicular Edge Computing and Networking for Intelligent Transportation)
Période du rapport: 2023-03-01 au 2025-02-28
Intelligent Transportation Systems (ITS) are vital for enhancing road safety, alleviating traffic congestion, and saving energy in transport. However, due to the complex and dynamic operating environments of ITS including fast-moving vehicles, fluctuating vehicular communications, and scarce computing resources, ITS face unprecedented challenges in meeting the stringent service requirements in terms of ultra-high reliability and low-latency, demanded by the emerging mission-critical applications (e.g. autonomous driving and real-time intelligent traffic control). To address these challenges, ASCENT aims to form an international, multidisciplinary, and inter-sectoral consortium with world-leading experts to create a novel Autonomous Vehicular Edge Computing and Networking system empowered by advanced Artificial Intelligence (AI) technologies towards achieving reliable and efficient ITS. Specifically, ASCENT will pioneer research and innovations (R&I) on ground-breaking technologies including: 1) a novel and scalable system architecture that enables agile and reliable ITS service provisioning; 2) an original distributed AI framework fuelled by bespoke federated deep learning methods to offer pervasive intelligence; 3) innovative analytics tools to accurately predict dynamic network status including network traffic and channel quality; 4) autonomous and smart resource management schemes to support mission-critical ITS services. ASCENT will boost the R&I capability of partners in multiple disciplines and create a long-term cross-discipline and cross-sector knowledge-sharing platform with complementary expertise. The researchers involved will be trained through extensive R&I actions and well-planned networking activities to enrich their skill sets as well as enhance their career perspectives. The outcomes of ASCENT will significantly contribute to enhance the EU’s competitiveness and transforming our transportation systems into safer, smarter, and greener future ITS.
ASCENT sets five coherent Research & Innovation Objectives (ROs):
RO-1: Design a novel, scalable and intelligent system architecture for VECN that can provide the efficient synergy, flexible interaction, and effective cooperation of VN and MEC empowered by AI technologies, and capitalizes on their unique capabilities to deliver ultra-high reliability and low-latency services for mission-critical ITS services. RO-1 will be measured by the performance and functionality of the VECN system architecture under different operating conditions.
RO-2: Create an original, ultra-reliable and distributed AI framework empowered by advanced deep learning algorithms that can significantly improve the system performance, communication efficiency, and training reliability of AI while enabling customized learning models for heterogeneous ITS services in VECN. RO-2 will be measured by comparison with the state-of-the-art AI frameworks in terms of reliability and efficiency.
RO-3: Design innovative online analytics tools based on multi-variate time-series data analysis technologies to accurately predict dynamic vehicular network status with respect to the network traffic and channel quality, which paves the way towards proactive and automated resource orchestration in VECN. RO-3 will be measured by comparing the accuracy of the proposed network status prediction algorithm with the existing algorithms.
RO-4: Develop autonomous and smart resource management schemes fuelled by advanced deep reinforcement learning technologies to orchestrate and schedule the computing, networking, and caching resources to meet the stringent performance requirements (i.e. low latency, high reliability and energy efficiency) of mission-critical ITS services. RO-4 will be measured in terms of latency, reliability, and efficiency of the developed resource management schemes.
RO-5: Establish a VECN system prototype that integrates the developed novel techniques and evaluate their performance in an ITS testbed with connected smart vehicles available in the project consortium to refine the developed technologies for VECN and further promote technological innovation and impact generation. RO-5 will be measured by the performance promotion of the developed prototype.
ASCENT sets five coherent Research & Innovation Objectives (ROs):
RO-1: Design a novel, scalable and intelligent system architecture for VECN that can provide the efficient synergy, flexible interaction, and effective cooperation of VN and MEC empowered by AI technologies, and capitalizes on their unique capabilities to deliver ultra-high reliability and low-latency services for mission-critical ITS services. RO-1 will be measured by the performance and functionality of the VECN system architecture under different operating conditions.
RO-2: Create an original, ultra-reliable and distributed AI framework empowered by advanced deep learning algorithms that can significantly improve the system performance, communication efficiency, and training reliability of AI while enabling customized learning models for heterogeneous ITS services in VECN. RO-2 will be measured by comparison with the state-of-the-art AI frameworks in terms of reliability and efficiency.
RO-3: Design innovative online analytics tools based on multi-variate time-series data analysis technologies to accurately predict dynamic vehicular network status with respect to the network traffic and channel quality, which paves the way towards proactive and automated resource orchestration in VECN. RO-3 will be measured by comparing the accuracy of the proposed network status prediction algorithm with the existing algorithms.
RO-4: Develop autonomous and smart resource management schemes fuelled by advanced deep reinforcement learning technologies to orchestrate and schedule the computing, networking, and caching resources to meet the stringent performance requirements (i.e. low latency, high reliability and energy efficiency) of mission-critical ITS services. RO-4 will be measured in terms of latency, reliability, and efficiency of the developed resource management schemes.
RO-5: Establish a VECN system prototype that integrates the developed novel techniques and evaluate their performance in an ITS testbed with connected smart vehicles available in the project consortium to refine the developed technologies for VECN and further promote technological innovation and impact generation. RO-5 will be measured by the performance promotion of the developed prototype.
The project has made significant progress in key technologies of AI and edge computing based ITS systems. The distributed AI algorithms, such as Federated Learning (FL), well fit to the vehicular networks. New FL algorithms, such as split FL and multi-task FL were proposed and showed excellent performance. Then reinforcement learning (RL) is also a good model for VENs.
Reliability in VNs are also developed including resilience and robustness. The traffic of VNs are models and trained based on advanced distributed learning algorithms.
The project team also applies meta-learning into offline DRL to learn a general agent under diverse content popularity for all edge servers, with the objective of achieving fast adaptation to the dynamic ITS environment while accelerating the training speed. Project ream developed a Digital Twin (DT)-driven task offloading framework for mobile edge computing (MEC), where DT is employed to map the MEC system into a virtual space and optimize the task offloading decisions. We proposed an agile edge cache replacement method based on offline-online deep reinforcement learning, to meet the dynamic requirements of mobile users in a MEC network.
In practical aspects, the project ream installed and deployed of the first 5G network for the VECN prototype where ITS applications was installed. This means the baseline for start building the end to end system prototype is ready at EXETER to input from other partners can be integrated to that system and get started with the integration ahead of the original plan.
Private Mobile Network with TJU university covering potential use case in manufacturing and using cellular network to recognize objects and performance.
For 4 EU beneficiary, a total of 45 PMs were planned for secondments during the first half project period. By the time of the Mid-Term Meeting, 32.4 PM secondment had been completed, resulting in an implementation rate of 72%.
Overall, the project has been completed as planned.
Reliability in VNs are also developed including resilience and robustness. The traffic of VNs are models and trained based on advanced distributed learning algorithms.
The project team also applies meta-learning into offline DRL to learn a general agent under diverse content popularity for all edge servers, with the objective of achieving fast adaptation to the dynamic ITS environment while accelerating the training speed. Project ream developed a Digital Twin (DT)-driven task offloading framework for mobile edge computing (MEC), where DT is employed to map the MEC system into a virtual space and optimize the task offloading decisions. We proposed an agile edge cache replacement method based on offline-online deep reinforcement learning, to meet the dynamic requirements of mobile users in a MEC network.
In practical aspects, the project ream installed and deployed of the first 5G network for the VECN prototype where ITS applications was installed. This means the baseline for start building the end to end system prototype is ready at EXETER to input from other partners can be integrated to that system and get started with the integration ahead of the original plan.
Private Mobile Network with TJU university covering potential use case in manufacturing and using cellular network to recognize objects and performance.
For 4 EU beneficiary, a total of 45 PMs were planned for secondments during the first half project period. By the time of the Mid-Term Meeting, 32.4 PM secondment had been completed, resulting in an implementation rate of 72%.
Overall, the project has been completed as planned.
The partners have actively collaborated and conducted research activities according to the scheduled plan. The partners have published around 26 high-quality research publications in the areas of artifical intelligence (AI), distributed learning, edge computing communications, networking, intelligent transport and the Internet of Things.