Skip to content

Visage Starts EU-Backed €3.1M Project On Trainee Stress

Two Croatian universities have joined face-tracking firm Visage Technologies to test AI that estimates learners’ stress and mental effort from cameras and sensors, starting with ship simulators.

Visage Starts EU-Backed €3.1M Project On Trainee Stress
Image courtesy: Visage Technologies

Visage Technologies, a computer vision company (software that interprets images) headquartered in Linköping, Sweden, has started a €3.1M research project to build AI that adjusts training to how stressed or mentally stretched a learner is. The EU is co-financing €1.9M of the cost.

Visage AI Labs, a team at the company’s Croatian arm, leads the work alongside the University of Zagreb’s Faculty of Humanities and Social Sciences and the University of Rijeka’s Faculty of Maritime Studies. The project runs from April 2026 to April 2029. Visage, founded in 2002, is best known for software that tracks faces, head movement and gaze, used in automotive, defence, industrial and healthcare applications.

Two Signals: Stress And Mental Effort

“The main idea behind the project is to help people learn by tracking two signals: stress level and cognitive load,” said Hrvoje Šamija, Visage’s chief business development officer. Cognitive load is the mental effort a person needs to take in information or carry out a task.

The team plans machine learning models that combine data from cameras, microphones, wearable devices and other sensors to estimate a learner’s state as they work. Results would then feed into the training itself, so lessons or exercises can change in real time.

Early targets are cognitive load, engagement, valence and arousal. Valence describes whether an experience feels positive or negative, while arousal describes how activated a person is, regardless of whether the feeling is good or bad. Visage says these are the initial focus, and the final set will depend on how reliably its models measure them once data is collected.

Three Products From One Platform

The project is set to produce three pieces of software. ACInsight SDK, a software development kit, is the core layer that merges signals from different sensors to estimate a person’s state. The other two products run on top of it.

ACTutor is designed to plug into existing e-learning platforms and adapt course material and the way it is delivered. ACTrainer is meant for professional training simulators, where it would check whether a simulated exercise puts trainees under pressure similar to the real job.

The first use case is training on ship-navigation simulators, the specialism of the Rijeka maritime faculty. At sea, mistakes made under heavy mental workload can contribute to accidents.

Human factors feature heavily in maritime safety data. The annual overview from the European Maritime Safety Agency (EMSA) found that 64.5% of accident events reported between 2015 and 2024 were linked to human action.

Edge Processing And EU Rules

Bruno Rosan, a senior R&D engineer at Visage, said the company is building the whole system in-house, from data processing and models to deployment on devices. “We can integrate any camera and any hardware, and everything we do runs on the edge,” Šamija said, meaning the data is processed on the device itself rather than in the cloud. Others are taking the same route. Video analytics firm Wobot has moved its camera software onto SiMa.ai chips that run on site.

The company says privacy, data protection and cybersecurity are built into the design. It names three EU laws. These are the General Data Protection Regulation (GDPR) on privacy, the risk-based AI Act and NIS2, a cybersecurity directive.

The AI Act bans AI that infers people’s emotions from biometric data in workplaces and educational institutions, with narrow exceptions for medical or safety reasons. European Commission guidance, following the act’s recitals, treats physical states such as fatigue as falling outside that definition. Visage says the AI Act’s requirements are built into the platform’s design. Its project materials do not detail how each measure, such as stress or valence, is treated under the act.

The project brings together Visage’s face-tracking experience, Croatian academic partners and EU funding to test whether software can read a learner’s state and adjust training in response. Its first proving ground is ship navigation, and the list of signals it measures remains open until the data comes in.

Add Morning Tick on Google