David Govedarevic Robotics & Software Engineer

With team TechSupport (HTL Wiener Neustadt) we won the European Conference on Educational Robotics, then travelled to the Global Conference on Educational Robotics in Norman, Oklahoma, where we placed 3rd in the top international bracket against roughly 47 teams from around the world.

• Developed the game strategy for a mecanum-drive bot built to solve the hardest — and highest-scoring — challenge on the entire game table.
• Wrote drum collecting & sorting algorithms used across both competition robots and supported both sub-teams with programming, code cleanup and debugging.
• Co-developed the robot with my teammate Amir Elmesiry — everything from hardware setup to the software stack was built by the students themselves.
• Trained and deployed an own vision model for autonomous drum detection on the field (see next project).

On the world championship I fully owned a second robot — a drum-collecting bot with a completely different design — including its AI vision system.

• Built a custom dataset (~600–700 photos of blue and pink drums on a white surface) and trained a YOLOv8n model, exported to ONNX.
• Programmed a DrumDetector class that starts inference on motion in the camera frames, detects drums and classifies blue vs. pink to drive the sorting mechanism.
• My first ML project — roughly 400 ms end-to-end inference on embedded hardware, and a hands-on lesson in what “production constraints” really mean.

Our team competes on the “Wombat” platform — a Raspberry Pi 3 at the core of a robot we set up entirely ourselves. We are currently replacing the official KIPR libraries (libwallaby) with a freshly written, in-house stack.

• Own architecture from low-level sensor & motor access up to a clean high-level mission API for competition code.
• Full hardware bring-up done by the students: OS setup, configuration, tooling and deployment with Git, Docker and Apptainer.

Together with a team I'm developing a realistic 3D racing simulator for RoboRacer teams — replacing the 2D simulators available today. Our own implementation on NVIDIA Isaac Sim with a 14-degree-of-freedom vehicle model tuned with real-world data.