Physical AI is part of various hardware systems – for example, robots and autonomous vehicles – but up until recently, AI specialists who could develop such solutions had been trained within separate subject areas: from robotics and mechatronics to more narrow ones, like design, programming, automatic control theory, and machine learning. The demand for a more integrated approach and interdisciplinary education has resulted in the creation of a united role-based competence model.
“Students who solely study machine learning are likely to face numerous difficulties when approaching real-world tasks. For one, when a robot bumps into an obstacle, it can’t brake sharply or pass through it like in a computer simulation – the dynamics of the systems needs to be taken into account. That’s why many companies today are on the lookout for those specialists who are equally good at algorithms and hardware and know how to launch a trained model on a device,” says Aleksei Vediakov, the deputy dean of ITMO’s Faculty of Control Systems and Robotics.
Designed to train sought-after physical AI specialists, the new model is based on a tool practical for both business and academia – namely, a description of current development roles. The model breaks down 12 roles across several fields: hardware, sensors, control algorithms, and AI. Among them are: an experimental researcher who generates and tests hypotheses, a developer who incorporates an algorithm prototype into robotic systems, a specialist who implements software and hardware solutions, and even a product manager who is meant to bridge the gap between the technological aspect of autonomous system development and business objectives.
Each role is accompanied by a set of competencies graded by skill levels – from basic comprehension to independent execution. That said, the model does not provide a fixed list of recommended courses or tasks; instead, it strictly defines the key fields of study and necessary proficiency levels. This allows universities specializing in engineering to add courses in machine learning and computer vision and those in AI – to expand their focus towards practical equipment and sensors.
The model serves as a foundation for Yandex Physical AI Garage, an educational program implemented at ITMO as a separate track within the Robotics and Artificial Intelligence Bachelor’s program. After their first year of studies, students of this track can compete for a spot in an advanced Yandex program tailored specifically for physical AI tasks. A total of 100 students will be enrolled in the program. Its fundamental studies will cover physics, mechanics, differential equations, and signal processing, while the practical classes will be led by Yandex experts. Each module from the fourth to the eighth semester will conclude with a month of intensive project activities – for example, students will need to develop navigation algorithms for an autonomous vehicle or environmental perception based on LiDAR and camera data. Students will finish the track with a portfolio of completed projects and experience of working in industrial teams.
