TRACK: Telemetry-Based Racing Analysis and Coaching Kit in Sim Racing Games
This study addresses the absence of an objective quantitative framework for sim racing driving behavior by proposing an unsupervised analytical framework based on telemetry data. Methodologically, a reinforcement learning reference agent is employed for independent normalization, mapping driving sessions into compact geometric fingerprints within a four-dimensional behavioral space to replace predefined labels. By integrating multidimensional feature engineering with rigorously statistically calibrated clustering algorithms, the framework enables driver profiling and performance evaluation. Experiments conducted on the Assetto Corsa Gym dataset reveal that corner types implicitly encode distinct behavioral dimensions, while demonstrating that only speed and consistency remain transferable across different vehicle configurations. These findings establish a robust analytical foundation for personalized driving instruction in simulated environments.