AWS DeepRacer
Developer Guide

This is prerelease documentation for a service in preview release. It is subject to change.

Build Your AWS DeepRacer Track

To build your own track to race your AWS DeepRacer vehicle and to achieve the best results, pay attention to the following factors.

  • Build your track to replicate the environment used in training. This is because models trained to follow track based on the road markings can exhibit different behaviors than models trained to follow the track based on other conditions or features.

  • Lay out your track using the same shape as the shape of the training track. For example, if you have uploaded to your vehicle a model trained on the Re:invent 2018 track, your track shape should look like this:

    
                    Image: An AWS DeepRacer track.

    The above track illustration doesn't show the walls surrounding the track.

  • Restrict the curb-side (turning) radius to the following limits:

    • At least 45 inches for corners smaller than or equal to 90 degrees

    • At least 55 inches for corners greater than 90 degrees.

  • Construct the track to have a smooth and uniform color surface of at least 30 inches wide and with minimal resistance.

  • Mark the track along the both sides with clear, unbroken and uniform-colored lines of 2-inch wide. The color of the markings should be of high contrast with respect to the track surface and barriers. Duct tapes have shown to be effective in some experiments.

  • Paint the off-track surface a color of sufficient contrast with respect to the on-track surface color.

  • For safety reasons, encircle the track with uniform-colored barriers of at least 2.5 feet tall and at 2 feet away from the track at all points along the track.

The following example track represents a scaled version of the track used at re:Invent 2018.


            Image: An AWS DeepRacer track.