AI Forum on the Use of Artificial Intelligence in Technical Practice

Together with Panasonic, we are developing tools that help design and manage production on the basis of data rather than estimates. At the core of the solution is a digital twin of the production system, which makes it possible to simulate operation in different configurations and to identify bottlenecks before any intervention on the real line. It is complemented by optimisation algorithms for workplace arrangement (facility layout) and by data-driven production optimisation, which uses operational data to find more efficient settings and plans. We will present the goals, the approach and the first steps of the collaboration, and show how NCUI research translates into concrete decisions in manufacturing.

Today an autonomous robot can move even in demanding terrain; the real challenge is its safe and efficient collaboration with people and other robots. Using the autonomous vehicle TAROS 6×6 and experience from the international exercises ARTEX and ALPEX 2026, we will show how we are moving from independent navigation to robotic systems that support shared tasks, such as supply. Within NCUI, we are developing methods for natural human–robot interaction. The aim is to create technologies for mutual perception between people and robots that enable meaningful collaboration in real-world environments.

The Biomedicine work package (WP6) uses artificial intelligence to link clinical and behavioural data for personalised prevention and population modelling. Its goal is a digital twin that predicts the individual and population-level development of health status from medical history, clinical parameters and a broad range of lifestyle data, and makes it possible to simulate the impact of interventions and their combinations (“what-if”). We will present the first step – a software platform for the secure collection and processing of sensitive health data, which OLTAIR Systems is developing with the Institute of Molecular and Translational Medicine, Faculty of Medicine and Dentistry, Palacký University Olomouc, and validating in a pilot deployment – and follow-on modules for continuous testing of cognitive functions and for evaluating data inputs from wearable electronics.

Every hospital laboratory is different: laboratories vary in sample volume, test mix, working habits and plans for the future. That is why even a laboratory automation system such as the DxA 5000 cannot have a universal form. Together with Beckman Coulter, we are developing a tool that turns laboratory design from guesswork into a precise, data-backed decision. At its core is a highly configurable digital twin of the DxA 5000 system. It can simulate laboratory operation in any configuration within moments and reveal bottlenecks before they arise. It is complemented by optimisation algorithms that decide where and how each test should be evaluated. By combining mathematical optimisation and machine learning, we look for configurations that are efficient today and will hold up as the hospital grows.

Artificial intelligence opens up new ways to predict the behaviour of energy systems and to plan and control their operation. The talk will present the research directions of the National Centre for Artificial Intelligence (NCUI) in energy, from forecasting generation and connecting energy sectors to decision support and system control. It will focus on how data and models can contribute to a better understanding of operation, the detection of inconsistencies and a more efficient use of resources. The individual topics are brought together by a view of AI as one piece of the energy puzzle, where research meets the needs of practice.

Language models have quickly become an integral part of modern information systems. In many cases, however, their application is very expensive and inefficient, and it is hard to update the knowledge held in the models. These limitations hold back some applications, for example in computer security. We are therefore developing small, inexpensive models that can easily be retrained for a new problem or domain and quickly updated to reflect new facts. To make this as simple as possible, we are developing an integrated tool that uses active learning to combine small, inexpensive models with large, expensive ones and distils large models into small ones, so that the user does not see a difference in use.