Marijo Alilović

I developed a methodology and algorithm for generating routes that maximize ADAS scenario coverage within user-defined boundaries, achieving significantly higher coverage than conventional methods. The tool enables highly optimized public road testing campaigns while minimizing cost.

I developed an algorithm for constructing compliant real-drive emission cycles, fully adaptable to any market through user-defined legislative inputs. The algorithm allows cycles to be optimized for low or high dynamics, allowing testing under extreme conditions.

I developed an automated daily fleet data reporting system featuring interactive tables and visualizations for easy KPI tracking and trend analysis. I also built a full data pipeline including import, cleaning, transformation, aggregation, and visualization.

In this project, my role encompassed the systematic data extraction from the database, followed by a meticulous tidying process to enhance its usability. Subsequently, I delved into an exploratory phase to gain deeper insights into the data's intricacies. Leveraging machine learning techniques, I navigated through various models to accurately predict the state of battery health. This iterative process allowed for a comprehensive understanding of the data landscape and facilitated the development of robust predictive models.

I developed a tool for enriching GPS data with diverse metadata via user-specified APIs. This enables distribution analysis and identification of conditions under which ADAS systems most frequently fail. All 1D, 2D, and country map distributions are represented using Plotly and Leaflet.

Developed algorithms to compute combinatorial Operational Design Domain coverage efficiently and to generate a multitude of gap-filling scenarios that maximize dissimilarity from existing scenario databases, enhancing test diversity and robustness.

Developed an algorithm to determine the feasible and optimal execution order of driving scenarios on a testbed, significantly reducing total test duration by minimizing idle time and unnecessary transitions.