Teaching an LLM the art of motor racing

Hello all,

when it comes to building an AI which can act upon huge amount of data with sufficient reasoning capabilities, it needs a lot of pre-processing most of the time. A very interesting area is understanding telemetry data of race cars. This data typically look like this:

The raw data is a huge collection of numbers for speed, throttle percentage, brake pressure, lateral and longitudinal G-forces, and so on, all sampled at a rate of 40 Hz minimum.

My goal was to give GPT 4o (mini) a full understanding of this data, so that the LLM can behave as a coach to give the driver suggestions and instructions how to changes his driver inputs and driving style for an upcoming corner to achieve better lap times.

Take a look at the following video. If you do not have time for the whole 10 minutes, skip to the middle of the video, where the on-track coaching starts.

youtu.be/mgfFkNh2_Lw

If you want to know more about the technical details of this implementation, let me know…