Ongoing work inside KA-RaceIng’s driverless environment, studying how perception, mapping and localisation become dependable vehicle behaviour under real-time constraints.
Trackside testing through a Formula Student cone course.Personal project archiveA second test run, filmed from the edge of the course.Personal project archive
The problem
An autonomous race car must estimate where it is and understand its surroundings quickly enough to act. The challenge is not a single model; it is the relationship between perception, localisation, planning and vehicle dynamics.
My contribution
01
Developing my understanding of the existing SLAM-based driving pipeline and the interfaces around it.
02
Exploring conventional and learned approaches to perception and localisation.
03
Working within a ROS 2 environment that connects sensor fusion, state estimation, planning and real-time components.
04
Using simulation and hardware integration as complementary ways to understand system behaviour.
Technical challenges
01
Balancing model ambition with the predictability required by a real vehicle.
02
Reasoning about failure modes when perception and localisation influence every downstream decision.
03
Building intuition across both software behaviour and the physical vehicle.
What I learned
01
Autonomy is a systems discipline: local accuracy does not guarantee useful global behaviour.
02
Simulation accelerates iteration, but physical integration exposes the assumptions that matter.
03
Comparisons between learned and conventional methods need careful, scenario-based evaluation.
Potential next steps
01
Deepen my contribution as responsibilities develop.
02
Compare approaches using clear robustness and real-time criteria.
03
Continue connecting perception outputs to the behaviour of the complete vehicle.
