Some games, like iRacing, don't provide the x/y coordinates in real-time. But we still get other telemetry, like SteeringAngle and Throttle and Brake input. We could create a similar approach, looking at the SteeringAngle, like already applied in https://github.com/b4mad/racing/blob/main/notebooks/fast_lap_analysis-v2.ipynb
But all these approaches have edge cases and need some tweaking of thresholds.
How about training a deep learning model?
The input features of the model are SteeringAngle, Throttle and Brake. The output is the layout of the track.
Training data can be created from those games that provide x/y coordinates and a collection of tracks with known good layouts. The layouts can be hand tuned.
goern
commented
1 year ago
Some games provide x/y coordinates of the car on the map. This can be used to plot the track. We use this map to split the track into sections. I.e. a connection of corners and straights. See https://github.com/b4mad/racing/blob/main/notebooks/track_sections_from_coordinates.ipynb
Some games, like iRacing, don't provide the x/y coordinates in real-time. But we still get other telemetry, like SteeringAngle and Throttle and Brake input. We could create a similar approach, looking at the SteeringAngle, like already applied in https://github.com/b4mad/racing/blob/main/notebooks/fast_lap_analysis-v2.ipynb But all these approaches have edge cases and need some tweaking of thresholds.
How about training a deep learning model?
The input features of the model are SteeringAngle, Throttle and Brake. The output is the layout of the track. Training data can be created from those games that provide x/y coordinates and a collection of tracks with known good layouts. The layouts can be hand tuned.