Multi-Access Edge Computing (MEC)

Multi-Access Edge Computing (MEC) is an advanced computing architecture that shifts cloud computing capabilities from centralised data centres down to the network edge. By deploying computing and storage resources closer to data sources—such as roadside locations and base stations—it delivers low-latency, high-bandwidth data processing to end devices locally, serving as a vital bridge between the 'edge device' and the 'cloud'.

If cloud computing acts as the 'brain' handling complex decision-making, edge computing functions like the 'spinal reflex'—executing rapid, split-second responses right where data originates, drastically slashing the latency and bandwidth load incurred from cloud data transmission.

Positioning and Value in Connected Vehicles

Within V2X (Vehicle-to-Everything) cooperative intelligent transport systems, MEC is typically deployed at roadside cabinets or base station edges, serving as an integrated hub for 'perception, computing, communication, and control'. Its primary value lies in overcoming three major bottlenecks of traditional centralised cloud architectures: transmission-induced latencies ranging from tens to hundreds of milliseconds, bandwidth congestion caused by massive raw data uploads, and computational bottlenecks during peak concurrent cloud workloads.

Take millimetre-wave (mmWave) radar point cloud processing as an example: synchronised data capture from multiple radars at a single junction produces massive real-time data streams. Transmitting all of this back to the cloud for processing entails multi-stage operations like filtering, denoising, clustering, and segmentation—consuming enormous bandwidth while introducing erratic latency jitter. By handling point cloud processing locally via MEC nodes, only structured feature data and situational insights need to be uploaded. This cuts bandwidth consumption by over 80% and compresses end-to-end processing latency to under 20 ms, satisfying the stringent real-time requirements of dynamic autonomous driving controls.

Functions and Technical Architecture

In vehicle-to-infrastructure (V2I) cooperative scenarios, MEC nodes typically deliver the following core capabilities:

• Sensor Data Fusion: Integrates multiple roadside sensors—including cameras, mmWave radars, and LiDAR units—to perform local spatiotemporal synchronisation and multi-source data fusion. This enables precise detection, classification, and trajectory tracking of moving road users (motor vehicles, non-motorised transport, and pedestrians).

• AI Inference Acceleration: Armed with dedicated GPUs or AI accelerator chips to handle heavy image/video processing and AI model inference, with typical roadside MEC hardware delivering computing power in excess of 200 TOPS.

• Low-Latency Information Distribution: Broadcasts fused, structured situational data (such as collision risk warnings and traffic signal phase timing) to nearby vehicles via low-latency V2X direct communications, closing the loop for vehicle-to-infrastructure cooperation.

Architecturally, MEC represents the intermediate tier in the three-level 'cloud-edge-device' ecosystem: the central cloud handles large-scale model training and macro scheduling; the edge executes real-time inference and instant response; and the terminal layer (onboard vehicle systems and roadside sensors) carries out raw data capture and foundational actuation.

Industry Application Scenarios

MEC deployment across connected vehicle ecosystems spans several key use cases:

• Real-Time HD Map Distribution: Roadside units collect live traffic conditions, offload computing loads via edge-based AI inference, convert findings into structured map updates at the MEC node, and broadcast them to local traffic, ensuring inbound vehicles receive up-to-the-minute high-definition mapping data.

• Large-Scale Continuous Vehicle Tracking: Across extensive roadside sensor arrays, the MEC framework enables continuous, real-time tracking for tens of thousands of vehicles simultaneously. Real-world tests demonstrate that across a 157 km highway stretch equipped with 1,777 sensors, tracking achieved an average longitudinal error of 2.14 metres, lateral error of 0.84 metres, and an end-to-end processing latency of below 340 ms.

• Roadside Perception and Warning: MEC nodes process roadside radar and optical camera feeds locally to identify vulnerable road users and dynamic hazards, evaluating threat levels against live traffic conditions before pushing proactive alerts directly to vehicles.

Important Note: MEC serves as the foundational telecommunications and computing infrastructure underpinning vehicle-to-everything (V2X) collaboration and autonomous driving. The reliability and real-time responsiveness of its data processing remain heavily contingent on network coverage and connectivity stability; hence, it is designed to augment, rather than entirely replace, an individual vehicle's onboard perception and autonomous decision-making systems.