Just as WMS systems set intralogistics in motion, "Physical AI" artificial intelligence with the physical world via sensors, active intervention through robotics, and plant control. "Stray attacks" here not only have a digital impact but also multiply the damage in the real world. Those using these systems must ensure safety.
Image: NTT Data / zVg
“Physical AI will become increasingly prevalent in the coming years, from autonomous machines and robotics to intelligent infrastructures. This also increases the responsibility to design these systems to be robust and secure from the outset”says Oliver Köth, Head of Development for the DACH region at NTT Data, a leading provider of AI, digital business, and technologyservices for companies. “Therefore, security cannot be added to physical AI as an afterthought; it must be an integral part of the architecture, training, and operation.”
Vulnerable sensors
According to Köth, Physical AI requires companies to rethink the security, robustness, and design of their systems . While classic control systems like programmable logic controllers (PLCs) are based on clear, deterministic logic, these applications operate in open problem spaces. They interpret unstructured sensor data, derive context-dependent decisions, and act in the physical world. This close integration of perception, decision logic, and action creates new attack vectors and necessitates enhanced security concepts. (...)
O. Koeth
Electronic and optical sensors form the basis for most physical AI applications, for which cameras, force sensors, or position sensors provide the necessary data. If these are manipulated or altered by external influences, the system can make incorrect decisions. In industrial environments, even seemingly trivial factors such as unusual lighting conditions, dust, or electromagnetic interference can directly affect the sensors. Furthermore, there is also the risk of targeted manipulation by criminal actors.
Compare measured values
To minimize risks, companies can, for example, use sensor fusion . This involves comparing data from multiple sensors. Comparing the measured values allows for the automatic detection of inconsistent or unusual processes. Additionally, redundancy of critical sensors, regular calibration, and real-time anomaly monitoring significantly increase the robustness of physical AI systems.
O.Koeth/klk.
The full NTT Data report is available here.
