Fig 1.
The figure shows the track division of the circuit Barcelona-Catalunya. The left part of the Figure shows the corner division. The right one shows Sector I (Blue), Sector II (orange) and Sector III (Green).
Table 1.
Overall performance statistics.
Fig 2.
Lap times by user, session and cluster.
Each dot represents a lap, the position in the user box depends on the session and the color depends on the performance level. The grand mean is plotted in black dashed line and the group means are plotted in dashed lines with group color.
Table 2.
Laptime statistics for performance levels.
Fig 3.
Users were assigned to a performance level according the average lap assignment. No user was assigned to a level WORST. Starting with session 3 no user was assigned to level BAD. (Left) Summary TLX, performing at EXCELLENT level reports lower Effort than GOOD and VERY GOOD. (Right) TLX per group and session illustrates apparent difference in reported Mental Demand and Frustration between EXCELLENT and VERY GOOD.
Fig 4.
Best lap times by performance level.
Fig 5.
Telemetry of all laps from VERYGOOD group (red) and EXCELLENT (cyan) group split into steering, brake and throttle.
Fig 6.
Block diagram of the features calculation process for predicting the lap performance based on the driving behaviour.
Fig 7.
Linear regression predicting the lap time of human drivers using selected features (left), histogram of prediction error (right) on the test set.
Table 3.
Absolute error in seconds (mean and std) for predicting the lap time and segment time based on the features extracted form segment(s).
Fig 8.
Comparison between a lap from the human group EXCELLENT (red) and a lap from the human group BAD (green).
The image shows the corner 1 of Barcelona. The X-axis represents distance through the racing line. The Y-axis (from top to bottom): Vehicle speed, steering wheel angle, throttle position, brake position, front left tire saturation, lateral acceleration, longitudinal acceleration.
Fig 9.
Comparison between a lap from the human group EXCELLENT (red) and a lap from the human group BAD (green).
The image shows the corner 3 of Barcelona. The X-axis represents distance through the racing line. The Y-axis (from top to bottom): Vehicle speed, steering wheel angle, throttle position, brake position, front left tire saturation, lateral acceleration, longitudinal acceleration.
Fig 10.
Comparison between a lap from the human group EXCELLENT (red) and a lap from the human group BAD (green).
The image shows the corner 10 of Barcelona. The X-axis represents distance through the racing line. The Y-axis (from top to bottom): Vehicle speed, steering wheel angle, throttle position, brake position, front left tire saturation, lateral acceleration, longitudinal acceleration.
Fig 11.
Comparison between a lap from the AI group EXCELLENT (green) and the AI group BAD (red).
The image shows the corner 1 of Barcelona. The X-axis represents distance through the racing line. The Y-axis (from top to bottom): Vehicle speed, steering wheel angle, throttle position, brake position, front left tire saturation, lateral acceleration, longitudinal acceleration.
Fig 12.
Comparison between a lap from the AI group EXCELLENT (green) and the AI group BAD (red).
The image shows the corner 3 of Barcelona. The X-axis represents distance through the racing line. The Y-axis (from top to bottom): Vehicle speed, steering wheel angle, throttle position, brake position, front left tire saturation, lateral acceleration, longitudinal acceleration.
Fig 13.
Comparison between a lap from the AI group EXCELLENT (green) and the AI group BAD (red).
The image shows the corner 10 of Barcelona. The X-axis represents distance through the racing line. The Y-axis (from top to bottom): Vehicle speed, steering wheel angle, throttle position, brake position, front left tire saturation, lateral acceleration, longitudinal acceleration.
Table 4.
Feature weights of the linear regression models showing how each feature affected the lap time prediction for human drivers (HD on the left) and RL agents (RL on the right).
Table 5.
Telemetry features used as input.
Fig 14.
Progress the algorithm while training.
The y-axis is the lap time and the x-axis is the total steps seen by the model. Brown dots are laps that fall in the “bad” lap group form Part I. Orange are the middle group. Gray are the good laps and red the excellent. Yellow and black are the best 50% and 25% of the excellent group, respectively. Right: the x-axis represents the number of simulated steps. Left: The x-axis represents the number of laps.
Fig 15.
Progress of the algorithm while training in each sector of the track.
Sectors I, II, III correspond to the right middle and left images. The x-axis represents the number of laps and the y-axis represents the time to complete the Sector.
Table 6.
Model performance across the different groups.
Fig 16.
Comparison between the best human lap (red) and the best result of the algorithm (green).
The image shows the Sector I of Barcelona. The X-axis represents distance through the racing line. The Y-axis (from top to bottom): Vehicle speed, steering wheel angle, throttle position, brake position, front left tire saturation, lateral acceleration, longitudinal acceleration.
Fig 17.
Comparison between the best human lap (red) and the best result of the algorithm (green).
The image shows Sector II of Barcelona. The X-axis represents distance through the racing line. The Y-axis (from top to bottom): Vehicle speed, steering wheel angle, throttle position, brake position, front left tire saturation, lateral acceleration, longitudinal acceleration.
Fig 18.
Comparison between the best human lap (red) and the best result of the algorithm (green).
The image shows the corner 10 of Barcelona. The X-axis represents distance through the racing line. The Y-axis (from top to bottom): Vehicle speed, steering wheel angle, throttle position, brake position, front left tire saturation, lateral acceleration, longitudinal acceleration.
Fig 19.
Linear regression predicting the lap time of the RL agent laps by training the model trained on the laps of human drivers.
Fig 20.
Linear regression predicting the lap time of RL agents based on the selected features (left) and the histogram of prediction error (right) on the test set.
Fig 21.
Scaled features of the laps performed by human drivers and RL agent.
The laps are grouped based on the performance levels. The features are scaled to have a mean set to zero and a standard deviation set to one.
Fig 22.
Speed channel comparison between all the laps from group EXCELLENT and all the laps from group VERYGOOD.
Fig 23.
Throttle position comparison between all the laps from group EXCELLENT and all the laps from group VERYGOOD.
Fig 24.
Steering wheel comparison between all the laps from group EXCELLENT and all the laps from group VERYGOOD.
Fig 25.
Brake comparison between all the laps from group EXCELLENT and all the laps from group VERYGOOD.
Fig 26.
Engine RPMs comparison between all the laps from group EXCELLENT and all the laps from group VERYGOOD.