Vice-President of Moscow Polytech spoke about the safety of unmanned transport and problems of artificial intelligence at the forum “TEKHNOPROM – 2023”.
The expertise of Moscow Polytech is based on the work with ground transportation vehicles. In recent years, the University has been developing several projects in the sphere of unmanned transport – these are pre-production samples of a minivan for parking areas Smart shuttle, the unmanned Housing and Utilities Platform, as well as a low-tonnage platform for autonomous systems.
All the developments of Moscow Polytech are equipped with intelligent control systems, which use neural network systems of processing information retrieved from sensors. However, along with the progress and innovations new challenges emerge. During the planning stage, much time was dedicated to the vulnerability of artificial intelligence itself. In particular, the situation of incorrect machine learning are possible, when wrong images are mixed into the data set.
This may lead to incorrect interpretation of road signs by the system or loss of control over a traffic lane.
Detection of such a substitution while learning is rather hard to implement, and it is hard to reveal. Hence, the issue of critical vulnerabilities of highly automated transportation means arises, such as wrong identification of road signs, losing control over a traffic lane, incorrect detection of objects.
“While talking about safety of unmanned transport, it is always necessary to understand the fundamental issue of reliability, safety and failure-free feature of artificial intelligence systems. These are algorithms of the unmanned transport interaction with surrounding world,” - noted Anton Nalivaiko.
Valentin Selifanov, deputy head of a Novosibirsk division of the JSC "Infotex", supported the scientist’s thesis with an example of the protection of unmanned vehicles in his report “Protection of Control Channels and Data Transmission of UAS”: “The main threat is the threat of information security. It must be ensured already at the stage of manufacturing unmanned devices, including inculcation, differentiation of access rights with authorization, trusted downloading, and trusted storage of data”.
Vice-President of Moscow Polytech concluded that the most evident and efficient measure to protect the safety of unmanned vehicles is normative and legal regulation in the area of the artificial intelligence development technologies. The effective measures also include the expert assessment and independent evaluation, validation and verification of data objects on which systems are trained.
