Page 803 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

define associated bandwidth requirements and uplink reliability thresholds to ensure
consistent communication performance.
• REQ-03: It is critical to differentiate between static data collection (for offline model
training) and real-time telemetry (for online inference), while also detailing spatial-
temporal data coverage expectations and data modalities including GPS, LiDAR, and Transport 4.10: Intelligent
RSSI.
• REQ-04: It is mandatory to define specific technical capabilities for the simulation platform,
including support for multi-radio vehicular mobility, stochastic channel modeling, and
integration with SDN/MEC emulators such as Mininet-WiFi, OMNeT++, or NS-3[6], to
accurately emulate real-world transport scenarios.
• REQ-05: It is critical to separate and articulate the requirements for dynamic machine
learning models, including support for online learning, explainability (via eXplainable
AI/XAI), model quantization for efficient edge deployment, and real-time adaptability to
domain shifts using techniques like transfer learning or continual learning.

4 Sequence Diagram

The workflow of the use case, as shown in Fig. 1, follows the sequence of tasks as shown in the
above sequence diagram, Fig. 2. Initially, the SDN module inside the base station aggregates
network data, including GPS information from the vehicles, and shares it with the clustering
engine and the pQoS engine, which are hosted at the MEC. The clustering engine performs
a stationarity test on the data, forms a vehicle cluster, and elects a leader vehicle (viz. Vehicle
1). The cluster data, including the number of vehicles and the leader vehicle's identity, is then
sent to the respective vehicles. And the pQoS engine predicts the network’s QoS and relays
this information to the leader.

If the QoS received by the leader meets the threshold, then the leader sends an offloading
request to the base station. The base station then communicates with the task offloading
manager, located near the leader vehicle. The task offloading manager executes these tasks in
the cloud and makes decisions regarding task offloading. The task offloading manager sends

767