Software and Sensor Issues for Autonomous Systems based on Machine Learning Solutions

The paper aims at investigating the most adequate strategies to develop efficient Machine Learning solutions for Autonomous Vehicles used for terrestrial, air, and maritime transportation. The development of systems with capabilities of performing some parts or even a full mission by adopting fully autonomous solutions has encouraged the development of new interfaces with human pilots. Standard interfaces are based on traditional systems that require low-level actions, such as the one related to steering and instrument monitoring. Indeed, autonomous transport systems have the capability to develop complex logic solutions to self-generate low-level actions. In this case, the role of the pilot is the development of high-level decisions rather than low-level steering of the vehicle. This condition permits the pilot to focus his workload on the most important issues related to driving the vehicle, thus reducing the risk of distraction determined by low- level steering. Therefore, the type and the layout of the developed interface will exploit recent technologies, such as touchscreens, voice recognition, and synthetic vision. The interaction looks like the one provided when the pilot relates to other humans rather than a machine. This condition is verified if the transport system is provided with Artificial Intelligence solutions based on Machine Learning. The paper discuss proper testing strategies to evaluate the adoption of specific interfaces in replacement of traditional ones. The final goal is to demonstrate the efficiency of the proposed testing strategies to select adequate interfaces that improve the quality and the safety of transportation. Four case studies are discussed to highlight efficient prototypical systems to be used in these applications.