AIthena D5.2 Report on initial use case evaluation

Deployment and testing of the AI-driven CCAM technologies

How do deployment and testing efforts work in AI-driven CCAM technologies? What is the value of deployment and testing efforts?

In the development of AI-driven technologies, training data for machine learning models is crucial to improve the efficiency and accuracy of the desired outputs of AI-based systems. Similarly, when it comes to Cooperative, Connected and Automated Mobility (CCAM) technologies, training data is derived from deployment and testing efforts when developing AI-enabled features and functions in vehicles.

@ AIthena D5.2

As part of the AIthena project, there are four use cases that are being explored to strengthen the accountability and robustness of AI-based CCAM. These four use cases are:

  • UC-1: Trustworthy Perception Systems for CCAM
  • UC-2: AI extended Situational Awareness/Understanding
  • UC-3: Trustworthy and Human understandable decision-making
  • UC-4: AI-based Traffic Management

For each use case, specific AI algorithms are used and tested to improve the performance of the desired outcomes programmed into the vehicle.

For example, in Use Case 1: Trustworthy perception systems for CCAM, AIthena research partner SIEMENS examined the safety architecture of autonomous vehicles to ensure that all safety features are working correctly. Researchers dissected the safety loop architecture built into the vehicle to observe how backup sensors, and diagnostic mechanisms interact to detect any failures. SIEMENS simulated a road environment with its Simcenter Prescan software, enabling researchers to understand how autonomous vehicles make their decisions based on what the vehicle sees or perceives in its environment.

By simulating different weather and lighting conditions, researchers can observe when a vehicle is most likely to succeed or fail in its intended autonomous operation. These observations can then be used to identify strengths, risks, and areas for improvement. In autonomous vehicles, identifying and correcting faults is critical to safety.

As another example, in Use Case 4: AI-based Traffic Management, researchers simulated the interactions of autonomous vehicles and non-autonomous vehicles on a modelled road with traffic, modelling the individual behaviour of other vehicles by adjusting their speed and manoeuvres.

Researchers have created a proof-of-concept design to test the interactions between autonomous vehicles and traffic management information through a microsimulation. The simulation is based on a single-lane urban road where a vehicle breaks down in the middle of traffic. The researchers found that when the AI-based autonomous vehicles receive information from traffic management networks, the vehicle is more likely to overtake the stranded vehicle.

By studying the results of these microsimulations, researchers can provide more explainable and transparent explanations for the decisions made by AI-based autonomous vehicles. These efforts then extend to the creation of more trustworthy AI-based traffic management systems.

“This deliverable is the second in WP5 and describes the initial use case evaluation for the first cycle of the project. The report concentrates on evaluating the results of the individual partners, while an evaluation of the joint work within the use cases is planned for the next deliverable (D5.3) at the end of the second cycle. The content is focused on the development of reliable perception systems for connected and automated mobility (CCAM). It explores various use cases, including pedestrian detection, semantic labelling, situation awareness, and vehicle guidance. Key topics covered are explainable AI, multi-modal perception, uncertainty quantification, and the integration of traffic management information. The goal is to enhance the safety and efficiency of autonomous vehicles by improving their ability to perceive the environment, make informed decisions, and interact with other road users.”

Bernhard Hillbrand, Virtual Vehicle (VIF) – Lead Partner of Deliverable 5.2 and leader of Work Package WP5 – DEPLOY & TEST: AI-based CCAM Deployment – integration on operation and end user validation.

To understand how AI models are deployed and tested in other use cases, read the full deliverable here AITHENA-D5.2-Report-on-initial-use-case-evaluation.pdf.