Original title: A Low-Cost, Open Platform for End-to-End Autonomous Driving on a Miniature Ackermann Vehicle
Gustavo Claudio Karl Couto, Eric Aislan Antonelo, Gabriel George Zipperer · September 2026
Testing self-driving cars on full-sized vehicles is expensive and difficult to do safely. To fix this, researchers built an affordable, open-source testing platform using a small, car-like robot that uses Ackermann steering. They also made a printed mini-city track, data tools, and a computer simulation version called a digital twin. To test it, they used behavior cloning, an artificial intelligence method that learns by copying humans. The system takes a picture from an on-board camera plus a driving command, then directly controls the car's steering and speed. In real driving tests, the mini car followed lanes and made turns with an average distance error from its route of 6.1 cm, very close to human drivers at 4.7 cm. The team also found that widening the camera view in the simulation improved driving, and mixing real driving data with translated simulation pictures helped the car successfully finish every track route.
Imagine you are learning to cook by watching a chef and then trying it in your own kitchen. The challenge is that your kitchen might look different, and you might not have all the ingredients or equipment. In this paper, researchers built a small car and a virtual copy of its track. By practicing in the virtual world and tweaking the images to look real, the car learns to drive all routes, much like how rehearsing with realistic props before an actual event builds confidence and skill.
This matters because autonomous driving research often requires expensive equipment, but this platform makes it accessible to more researchers. By providing open tools and a digital twin, it allows for controlled experiments that can lead to better real-world self-driving systems.
The abstract does not explicitly state limitations, but the work is focused on miniature vehicles and a specific urban track. It uses baseline methods like behavior cloning and sim-to-real translation, which may not generalize to all autonomous driving scenarios.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
A Low-Cost, Open Platform for End-to-End Autonomous Driving on a Miniature Ackermann Vehicle — Gustavo Claudio Karl Couto, Eric Aislan Antonelo, Gabriel George Zipperer
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.