진학프로석·박사 채용 접수 플랫폼

Czech Technical University in Prague (CTU) offers a fellowship program, the CTU Global Postdoc Fellowship. This new and attractive two-year fellowship-program offers excellent researchers who have recently completed their PhD the chance to continue their research career at CTU. Fellows receive a two year fellowship and become members of a team led by a mentor.

The scholarship applicant must meet the following conditions on the date of application:

• • be no more than 7 years since obtaining the first Ph.D. degree,
• • Ph.D. studies at a university outside the Czech Republic or have completed at least a one-year working research stay abroad (outside the Czech Republic),
• • be an author (co-author) of three or more publications in a journal with IF or CORE A*/A conference paper.

Robust machine control assumes modeling of robot-environment interactions. An example may include an outdoor autonomous ground robot that needs to be aware of its model and how the terrain will interact with it when a control sequence is executed. A flying robot may benefit from knowing the wind field ahead to model aerodynamic forces correctly.

However, building robust perception systems that can efficiently adapt in a self-supervised manner to novel environments remains a significant challenge. We identify three core issues: (i) black-box models that ignore the robot's physical embodiment suffer from poor generalization, weak explainability, and limited transferability; (ii) sample-inefficient learning requires large volumes of annotated, domain-specific data; and (iii) complex architectures with tightly coupled components hinder modular adaptation. To address these limitations, we research a physics-guided machine learning framework that integrates physical knowledge with data-driven methods. Physical knowledge includes, among others, kinematics and the dynamics of the robot, terrain interaction and contact models, and environmental physics, such as wind. The physics can be incorporated in various ways. Two methods now researched most intensively are i) trainable machine learning pipelines may embed differentiable physical models, and ii) the learning process may be informed by constraining the predicted variable to obey physical laws; we can see it as physics-informed losses. We will seek new ways to embed the physics.

Our approach aims to enable explainable, embodiment-aware, and probabilistically consistent adaptation from onboard sensory data via end-to-end differentiable architectures, enhancing robustness, efficiency, and generalization across diverse robotic platforms and environments.

Group and supervision: Research will be conducted within the Vision and Robotics Group. The group has extensive experience with real robotics, such as successful participation in the DARPA SubT challenge (https://robotics.fel.cvut.cz/cras/darpa-subt/) and several state-of-the-art robotics platforms and sensors (https://robotics.fel.cvut.cz/cras/robots/). The Department has access to a high-performance computational cluster dedicated to artificial intelligence research and developments using traditional multi-CPU systems, but also GPUs.

Skills/Qualifications

We seek highly motivated applicants with a PhD in robotics, AI, or related fields and a proven track record relevant to the topic - publications in top journals or conferences (e.g. computer vision (CVPR/ICCV/ECCV), machine learning (NeurIPS/ICML), or robotics (ICRA, IROS, RSS, CoRL; IEEE-TRO, IJRR).