Send a link

Machine Learning

The goal of the Machine Learning process is trying to improve the track parameters that have been found manually by trial and error. We can enable the machine learning process by selecting the corresponding option from the Pro Racing AI menu:

After that, we can build a stand-alone build and leave the AI running alone, trying to improve its lap times, on a spare computer. Then we will see the machine learning configuration screen:

  • Number of Laps: once we configure all parameters, the AI will start driving alone and testing different values for the parameters in each sector of the track. After completing the selecter number of laps, it will compare the current lap times with the best lap times achieved until now, to see if the new parameters manage to improve the lap times or not. if we select a small number of laps, the machine learning process will converge faster but the solution obtained by the process may be sub-optimal, so It's recommended to use a value equals or greater than 3.
  • Simulation Speed: this parameter will allow us to select the timeScale that will be used during the machine learning process. High simulation speeds will accelerate the convergence time, but they will decrese the physical simulation accuracy , so this parameter must be used carefully. The fps counter shouldn't go below 30-50 fps because the quality of the physical simulation could be degraded too much.
  • Margin of Error: this value is used for calculating the new values for track sector parameters. Low Margin of Error values can make the AI get trapped in a local minima while too large values can make the machine learning process no converge at all.
  • Track Information File: absolute path to the file where the track information is stored.
  • Save Changes to File: absolute path to the file where the new track parameters will be saved.

After filling all the information, we can press the Start Trial button to start the machine learning process.

Finally, the AI will start racing alone while we can see the lap times and the values used for the different parameters. Whenever the AI manages to improve its lap times, it will save the new track information into the selected file, so we can stop this process safely at any moment, there is no risk of losing the changes made by the AI.