What is it about?
Here a minuture car learns to stay on the road only based on what the frontal camera sees. The car is first trained in simulation, whereafter the learned behavior is transfered to a miniture highway in our lab. Essential here is that during training not only the perfect trajectory is shown, but also the situations where the car drifts off. The less perfect situations help to learn to recover once the car has to drive in the real world.
Featured Image
Photo by S. Tsuchiya on Unsplash
Why is it important?
There are many datasets from autonomous driving companies, which can be used to learn the perception part. Yet, to learn the control of a car you need a smaller car, which is more affordable and where crashes are non-destructive. The combination of a Nvidia Jetracer on a Duckiebot highway is a perfect trainingground, as this work demonstrates.
Perspectives
Both Nvidia and the Duckiebot initiative are great in their efforts to combine education with research, which accessable hardware, combined with extensive documentation, tutorials and webinars. The combination of those two initiatives is even better, allowing to experiment with autonomous driving in a safe setting.
Dr. Arnoud Visser
Universiteit van Amsterdam
Read the Original
This page is a summary of: Learning to Drive Fast on a DuckieTown Highway, January 2022, Springer Science + Business Media,
DOI: 10.1007/978-3-030-95892-3_14.
You can read the full text:
Contributors
The following have contributed to this page
