Dzik Autonomous Racing

About

Fully autonomous

“If everything seems under control, you're not going fast enough.” Mario Andretti had it right for humans. For an autonomous car, it flips: if you don't control everything, you can't go fast enough.

It started in 2018 with the Udacity Self-Driving Car Engineer nanodegree and the itch to put theory on real asphalt. Our first car was a stock Erle Robotics kit on ArduPilot, enough to learn, not fast enough to race. So we built our own.

The platform is rebuilt for every competition we enter. The sensor stack is chosen again each time: what worked, what didn't, what's worth testing next. The algorithms have walked the full path, from a viciously tuned follow the gap, to manual trajectory planning on a map, to fully automated planning, to the ruthlessly optimized, tightly coupled loop of sensing, localization and control that the cars run today.

Every frame, every filter, every line of code exists for one reason: to push how fast a small car can go when the driver is an algorithm and the margin is a millisecond.

Until late 2024 that meant F1TENTH, now RoboRacer, indoors at robotics conferences, where the opponent is another team's algorithm. Six podiums in, we changed the question. The car moved outdoors onto asphalt, into the 1/8 class humans actually race, and the benchmark became a professional driver with a transmitter in their hands.

Team

Łukasz
Sztyber

Monika
Sztyber

Cars

Dzik1 | Dzik2

Engineered to the edge. Autonomous when we let them.

Dzik race car seen from the front at the start line of an outdoor asphalt track

Two Dzik cars on an indoor track with their rear lights lit

Dzik race car from the front left, Hokuyo LiDAR and cameras on the front deck

Chassis

Platform: Sworkz S35-GT2E
Length: 470 mm
Width: 300 mm
Height: 150 mm
Wheelbase: 335 mm
Frame: Custom carbon fiber, CNC cut (1 to 3 mm)
Weight: 3800 to 4200 g (depending on configuration)

Compute

Computing unit: NVIDIA Jetson Orin NX 16GB

Sensors

Wheel odometry: Highly accurate custom solution
IMU: Xsens 630R · Xsens Sirius · Analog Devices ADIS16475 · Safran STIM300 · Exail UMIX-40
Magnetometer: PNI RM3100
Depth & RGB cameras: Stereolabs ZED X One S (Narrow · Wide · Fisheye) as virtual stereo sets
LIDAR: Hokuyo UST-10LX · Hokuyo UST-30LC · RoboSense Airy 96

Remote control

Transmitter: Flysky Noble NB4+
FPV: DJI O4 Air Unit Pro + DJI Goggles 3

Achievements

Track record

Human vs AI · since 2025

The opponent is no longer another team's algorithm. It's a professional human driver, indoors or out, in the same 1/8 RC class. On lap time, the car is already on par.

More sessions