Projects
Research projects, competition entries, and open source work β from unsupervised LiDAR perception to autonomous robots competing in the field.
Physics-Guided Learning for Autonomous Navigation
PhD dissertation at CTU VRAS Group. Developing control algorithms for autonomous mobile robots that leverage terrain properties obtained through perception and physics-informed neural networks, enabling adaptive navigation across diverse outdoor environments.
FlowSeg4D: Unsupervised 4D Panoptic Segmentation
Master's thesis. An unsupervised framework for online 4D panoptic segmentation of LiDAR driving scenes that combines scene flow estimation with clustering to achieve temporal consistency without labeled data. Achieves LSTQ 46.9 on SemanticKITTI, rivalling supervised methods.
Outdoor Autonomous Robotics
Research project at CTU VRAS Group covering semantic segmentation, navigation, and localisation for robots operating in unstructured outdoor environments. Involves deployment on NVIDIA Jetson platforms and validation on HPC infrastructure.
Autonomous Road Crossing with a Mobile Robot
Bachelor's thesis. A behavior-tree control algorithm that lets a mobile robot safely cross roads by detecting and tracking surrounding vehicles. Evaluated in simulation and on real hardware. Received the Dean's Award at CTU FEE.
OSM data for autonomous robotics
A Python package for downloading and processing OpenStreetMap data for use in autonomous robotics applications. Provides tools to extract road networks, building footprints, and other relevant map features, and convert them into formats suitable for robot navigation and planning. Comes with a visualization tool to inspect and edit the extracted map data, perform high-level route planning, and export to ROS-compatible formats. ROS2 version includes a Tracker mode for real-time robot localization, state verification and control. Deployable as a ROS2 package and standalone Python library.
Online Robot Monitoring
Software for online monitoring of deployed autonomous robots β tracks position, sensor health, process states, and key metrics in real time. Server side processing and visualization application, and ROS2 node for data collection and transmission. Comming soon under an open source licence.
Robotour 2025 β 2nd Place
International outdoor robot competition. Our robot autonomously navigated an urban environment to deliver beer, placing 2nd among international teams. The software stack included ROS2-based navigation, sensor fusion for localization, and a custom behavior tree for task execution. The robot successfully completed the delivery route while avoiding pedestrians and stayed within the designated path and time limits.
Porsche Engineering Student Contest β 2nd Place
Developed full autonomy software for a 1:10 scale vehicle capable of high-speed circuit navigation, road sign recognition, and real-time obstacle avoidance. Competed against teams from European universities in accuracy and speed challenges.
RobosoutΔΕΎ β 1st Place
National maze-solving robot competition. Designed and programmed an autonomous robot capable of line following, pathfinding, and return navigation, achieving the fastest completion time through optimised control algorithms and sensor fusion.