Vladimír Petrík, a researcher in the Intelligent Machine Perception team at the Czech Institute of Informatics, Robotics and Cybernetics (CIIRC CTU), has been awarded a prestigious JUNIOR STAR grant from the Czech Science Foundation (GA ČR). His five-year project, PRINCIPIA, will focus on developing robotic intelligence that enables robots to learn new skills from online videos and improve them through safe, autonomous experimentation. Petrík was selected from 316 project proposals, joining 35 researchers awarded funding for 2027. He is also one of four successful applicants from the Czech Technical University in Prague (CTU).
Teaching Robots to Learn Safely from Their Mistakes
Today’s artificial intelligence systems can improve through vast amounts of trial and error. In digital environments, they can make mistakes repeatedly without causing real-world damage. For robots, however, the situation is more complicated. A single incorrect movement can damage an object, break equipment, or even put a person at risk.
This is why teaching robots new skills remains challenging and often requires direct human guidance. As a result, robots are still primarily used in controlled industrial environments, while their wider adoption in homes, hospitals, and everyday workplaces is limited, in part, by their inability to respond safely to unexpected situations.
Vladimír Petrík aims to overcome this challenge through his project, PRINCIPIA: Scaling Up Physical Robotic Intelligence with Newtonian Compliance.
At the heart of his approach is the ability of robots to adapt their behavior to physical conditions and interactions with their surroundings. For example, a robot should be able to recognize when a task requires greater force and when it needs to yield. While tightening a screw requires sufficient stiffness, unexpected contact with a person should trigger a compliant and safe response.
The goal is to make safety an integral part of robotic learning. This would allow robots to improve through their own experiments without every mistake posing a risk of damage to their surroundings.
From Online Videos to New Robotic Skills
If robots can learn to handle their own mistakes safely, they will be able to draw on a much wider range of information sources. One promising resource is the vast collection of instructional videos available online, which could help robots acquire new manipulation skills.
Visual information alone, however, is not enough. While cameras capture movement, they cannot directly measure the forces involved in physical interactions. The project will therefore use physics-based simulations to estimate these forces, enabling robots to learn even from incomplete or imperfect information.
The next step is autonomous experimentation. Robots should be able to identify what they have yet to master, design an appropriate experiment, carry it out safely, and gradually improve their abilities based on the results, without requiring constant human supervision.
Building on Years of Research at CIIRC CTU
The PRINCIPIA project builds on long-term research in robotic intelligence and machine perception at CIIRC CTU. Vladimír Petrík emphasizes that the successful grant proposal also reflects close collaboration with colleagues and a supportive research environment.
“A proposal like this doesn’t come from one person alone. I would like to thank the entire IMPACT team at CIIRC CTU for creating a research environment in which these ideas could mature, and especially Professor Josef Šivic for his long-standing support and feedback on the project proposal,” says Vladimír Petrík.
One of the project’s main scientific objectives is to address a fundamental limitation of today’s embodied artificial intelligence. Although modern robotic systems can plan actions based on visual information and language instructions, they often lack a sufficient understanding of the physical forces that determine whether manipulation will be successful and safe.
“In five years, I would like to demonstrate a robot that can learn a new everyday skill from a video and improve it on its own. We also plan to make everything openly available so that other researchers can build on our work,” Petrík adds.
Beyond its scientific findings, the project will provide the wider research community with open-source software, datasets, and trained models that other research teams will be able to use.
Bringing Robots into Everyday Life
The project’s results could eventually help bring robots beyond traditional industrial environments. The ability to learn safely, adapt to unfamiliar objects, and respond appropriately to physical contact with people is essential for their future use in homes, healthcare, logistics, and manufacturing environments where humans and robots share the same workspace.
The project will also contribute to a deeper scientific understanding of how artificial intelligence systems can gain experience through direct interaction with the physical world.
JUNIOR STAR Supports Outstanding Early-Career Researchers
The JUNIOR STAR programme of the Czech Science Foundation (GA ČR) supports outstanding early-career researchers who received their Ph.D. no more than eight years ago and have already achieved international scientific recognition. Its aim is to enable talented scientists to establish independent research groups and pursue ambitious research projects.
Each project can receive up to CZK 25 million in funding over five years.
For the 2027 call, the Czech Science Foundation received a total of 316 JUNIOR STAR project proposals, of which only 35 were selected for funding. This represents a success rate of approximately 11%. Together with 16 projects supported under the EXPRO programme, a total of 51 excellent research projects will begin in 2027.



