Page 799 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

contributing to crash prevention, enhancing vehicle safety, and uninterrupted communication.
However, proactive decision-making in such systems presents several challenges:

1. Mixed Traffic Conditions & Multi-Environment Scenarios: Roads are unpredictable with
human-driven vehicles, pedestrians, and unexpected obstacles influencing the platoon’s Transport 4.10: Intelligent
stability. Network link conditions are highly impacted by multi-environment scenarios such
as highways, avenues, residential areas, and park, etc.
2. Pre-programmed ML Models: Most of the applications today use pre-programmed ML
models, which are designed for specific scenarios, limiting their adaptability to diverse
mobility conditions and varying network environments.
3. Computational Costs: The cost of processing vehicular sensor data on board varies widely
depending on factors such as model complexity, the number of sensors used, processing
power required, and hardware capabilities, etc. While basic processing can cost only
a few cents per gigabyte, advanced computing needs, such as those for autonomous
driving, can reach tens of dollars per gigabyte, posing a significant challenge for real-time
QoS predictions.

Need for stationarity check and predictive QoS for vehicle safety:

Our use case focuses on developing a cost-effective decision-making system, ensuring safe
mobility for vehicle platooning in a V2X communication network. It leverages real-time data
exchange among vehicles, infrastructure, and networks. By utilizing the extensive data available,
we enable proactive decision-making that satisfies End-to-End (E2E) QoS requirements.
Proactive decision-making is based on a time stationarity check of multi-environment data
and predictive QoS (pQoS), e.g., latency, packet loss ratio, or throughput [2].

The time stationarity is a process that checks for a change in the unconditional joint probability
distribution of wireless network data with time. The pQoS is a metric that enables forecasting/
predicting wireless network quality well in advance.
• The stationarity check information is used for the formation of the optimum platoon,
ensuring safe distances between vehicles, and improving overall traffic efficiency.
• The E2E pQoS predicts the E2E quality of wireless connectivity of the communication
link. The E2E pQoS takes care of task offloading decisions. Without pQoS, any offloading
strategy would eventually result in service disruption.

By utilizing lightweight algorithms and edge computing, the system ensures affordability
while improving road safety [3], cost-efficiency, and autonomous coordination in connected
transportation environments.

Objective: To develop low-cost and safe mobility, End-to-End (E2E) pipelines for the reliable
prediction of QoS parameters and task offloading for vehicle platooning in multi-environment
scenarios by leveraging data from both mobile networks and vehicle sensors.

Use Case Status: The use case is part of a larger research project�

Detailed solution proposal

We address the challenge of unavailable or computationally expensive network data for QoS
prediction by enhancing our Machine Learning (ML) model with prior information from a
leading vehicle [4].

By analyzing correlations among vehicles traveling the same path, we observe that they
experience similar large-scale fading effects, leading to strong correlations with the leading

763