Inherited
Drive-by-wire actuation and control, the camera and LiDAR hardware, and the Apollo integration with GPS/IMU and the CAN bus—built for Providentia between 2016 and 2021.
Fortuna / full-scale research platform
A street-legal research vehicle.
When I started working with Fortuna in 2023, the vehicle had been inactive for more than a year. I took responsibility for bringing it back into operation. This involved reconstructing parts of the existing architecture and operating procedures, restoring the software and hardware components needed for autonomous driving, and gradually making the vehicle usable for research again.
Today I lead the technical work around the vehicle. We use it for research on autonomous-driving testing, simulation, and real-world validation. Several students and researchers contribute to the platform and use it for their own work.
2023Recovery and reintegration began
48h+Real-world driving in the 2026 perturbation study
MultipleModular and end-to-end ADS architectures
2Vehicle scales—a full-size Fortuna and a small-scale Donkey Car
Inside the vehicle
Two ways to drive the same car.
The driving stack can be an end-to-end model or a modular perception-planning-control pipeline. Bring your own laptop or PC, up to 1000 W, connect it over Ethernet, and start driving.
The platform
The Fortuna platform
Fortuna is fortiss’s full-size autonomous-driving research vehicle: a retrofitted, street-legal Volkswagen Passat Variant GTE with five cameras, three LiDARs providing 360° coverage, precise localization, onboard compute, and real-time access to steering and acceleration through the vehicle gateway.
We use it to run the same research question across offline models, hardware-in-the-loop setups, and a complete moving vehicle, then compare where the answer changes.
Inherited, recovered, changed
What I actually did.
Recovered
In 2023, after more than a year of inactivity: reassessed every subsystem, restored communication paths, and made the vehicle inspectable and operable again.
Changed since 2023
A ROS bridge alongside Apollo so any laptop or PC can drive the car over Ethernet, modular and end-to-end driving stacks, simulation integration, hardware-in-the-loop infrastructure, and the small-scale Donkey Car platform.
Recovery → research
From recovery to research.
Fortuna wasn’t built for me. It was fortiss’s automated-driving prototype for Providentia—then the project ended, and the car sat unused.
2016–21
Built for Providentia
As fortiss’s 5G-connected automated-vehicle prototype, Fortuna gathered data alongside roadside sensor infrastructure for Providentia, a project researching connected, automated driving on Germany’s A9 digital motorway test bed near Munich. When the project ended in 2021, so did the car’s job.
2023
Understand and recover
I assessed each inactive subsystem, restored communication paths, and made the vehicle inspectable and operable again.
2024–25
Rebuild it modular
Rather than restore the original single-purpose setup, I rebuilt Fortuna as a modular, ROS-based platform: the driving stack can simply be swapped, and so can the compute behind it—any laptop or PC up to 1000 W, connected over Ethernet. With students joining the project, we integrated modular and end-to-end driving stacks, sensor pipelines, and simulation.
2026
Test at full scale
For our ASE 2026 study, I compared 72 camera and LiDAR perturbations across offline analysis, hardware-in-the-loop execution, and closed-loop driving on Fortuna: more than 740 experiments and over 48 hours of autonomous driving.
NEXT
Turn fieldwork into shared infrastructure
I keep connecting physical drives to reproducible scenarios, open tools, student projects, and research questions the next team can build on.
System anatomy
Sensors, compute, and control.
- 01 Cameras for visual perception and end-to-end driving
- 02 Three LiDARs for 360° geometry, obstacles, and modular stacks
- 03 RTK localization and onboard ROS compute
- 04 Real-time drive-by-wire actuation and control
Hardware and logic
Hardware and distributed computing pipeline.
The original platform ran on Apollo alone. I expanded it into a distributed system: camera and LiDAR drivers publish over ROS from an Nvidia Drive PX2 and a dedicated PC, Apollo still owns GPS/IMU, the CAN bus, and the real-time control interface that talks to the dSpace MicroAutoBox II, and everything meets over Ethernet. A bridge machine runs the ROS master alongside an Apollo/ROS adapter, so a separate “intelligence” node (the system under test, plus whatever extra tooling a given experiment needs) can subscribe to sensor data and publish control commands without touching Apollo directly.
The small-scale platform
Donkey Car, Fortuna’s little brother.
Fortuna has a smaller sibling: a 1/16-scale Donkey Car with its own small-scale lab and a mixed-reality framework built around a digital twin of that lab. For our ASE 2025 study, we used it to compare simulation, vehicle-in-the-loop, and mixed-reality testing against real driving—vehicle-in-the-loop reduced the actuation gap, and mixed-reality testing reduced the perception gap, compared to simulation alone.
Read the ASE 2025 paper
Not a one-person project
Students who have driven this forward.
I lead the technical work around the platform under the supervision of Prof. Andrea Stocco, but the research around it is a shared effort. Here is some of the student projects that have grown around the platforms.
Davide Jannussi
Davide built Cam2Sim, turning our recorded drives into playable, closed-loop CARLA scenarios.
Xun Ma
Xun’s thesis studied how to adapt a model trained purely in simulation so it holds up in the real world.
Matthias Weil
As a bachelor’s student, Matthias co-led the real-world perturbation study that became our ASE 2026 paper.
Mirena Flores Valdez
Her master’s thesis became the ASE 2025 study comparing simulation, vehicle-in-the-loop, and mixed-reality testing on the small-scale platform.
Skander Jneyeh
For his bachelor’s thesis, Skander built a physical, IoT-connected traffic-light system for the small-scale platform.
Hannes Leonhard
As a bachelor's student, Hannes built PerturbationDrive, the image-perturbation framework for testing automated driving assistance systems.
Constantin Carste
Constantin's bachelor's thesis used Stable Diffusion and CARLA to reconstruct real driving scenarios and generate matching road layouts from recordings—the basis Cam2Sim was later built on.
Timo Götz
Timo's industrial thesis, done in collaboration with BMW, automatically validates acoustic test-bench recordings of electric motors.
Gökce Sengün
Gökce's thesis extends the ICST 2024 quality-metrics study to system-level testing on a physical vehicle.
Several of these students came through a practicum course I help organize on testing autonomous driving systems. More on teaching
Just for fun
I also scanned Fortuna into a 3D model.
You can take it for a lap around the track, no research question required.
Open it up