What is AG-Drive?
AG-Drive provides an interactive and accelerate AI-driven framework for Practical driving courses and driving license exams.
The framework leverages on 5G network, coupled with a custom AR tool able to offer real-time feedback on driving behavior through multiple-sensor data collection and processing in real-time. The distributed application will be able to perform driving behavior classifications and will suggest specific improvements, targeting driver’s gaze, body posture, vehicle position, statistics and trajectory.
AG-Drive also targets the creation of a large dataset for driving tests classification of key performance parameters. The system is designed to represent a useful tool for operators, helping with all the certifications, driving licenses and so on.
A demonstration is available at the Track Testbed Premises located in Pontedera (Pisa, Italy).
AG-Drive Components and Infrastructure
The AG-Drive system is represented by a distributed framework able to collect and process data in real time from multiple sensors. Sensors involved are:
Gaze-tracking glasses
Cameras
Inclination sensor
OBD port
Posture sensors
Low-latency is granted by a private 5G solution employed in the driving track testbed will serve as the backbone for connecting these sensors, processing collected data and provide feedback to drivers and instructors.
AG-Drive is deployed in a pilot infrastructure consisting of an innovative motorcycle maneuvering area in the city of Pontedera (Pisa), regularly used for professional motorcycle training and exams.
The area includes a building already served with internet connection and Wi-Fi and hosting servers for edge computing and networking. The area is already equipped with 9 video cameras. Their placement has been carefully designed to effectively cover all relevant angles of the maneuvering area. The area is also equipped with private 5G connectivity.
Related Projects
AI-RIDE
The AI-RIDE project uses artificial intelligence to train motorcyclists through a series of sensors installed on the helmet, the motorcycle, and trackside cameras. The system highlights mistakes and incorrect behaviors during riding by detecting the motorcycle’s position, the rider’s path, and the correct positioning of the gaze while driving.
During the exam, the system can automatically detect errors along the predefined route, impacts with boundary cones, and track departures.
The research has enabled these checks to be carried out using neural networks trained and optimized to work with hardware that can be quickly installed in any driving school.
5G-SDA
Private 5G network for low-latency, time-critical, and high-bandwidth driver training applications.
The 5G-SDA project aims to provide the FIDAL network with a pilot infrastructure — a testbed on track for driver training and examinations. It is an AI-assisted performance evaluation tool that leverages the 5G network, combined with a new augmented reality system offering real-time feedback on driving behavior.
In particular, it focuses on gaze direction tracking using smart glasses and other sensors, enhancing driver training, infrastructure management, traffic control, and road safety. The goal is to benefit vulnerable road users and reduce accidents.