64 million simulation steps recorded from Assetto Corsa include 2.3 million human-driven steps and the remainder from Soft Actor-Critic (SAC) policies. The data covers 15 drivers across professional, expert, casual, and beginner skill levels completing at least five laps per track and car combination.
Use Cases
- Train imitation learning agents using the human-driven steps to model different driver behaviors
- Evaluate offline reinforcement learning algorithms using the 61.7 million SAC-generated steps
- Perform cross-skill analysis by comparing trajectories from professional e-sports drivers against beginners
- Develop autonomous racing agents capable of generalizing across different tracks and car configurations
Strengths
- 64 million total simulation steps for autonomous driving research
- 2.3 million human-driven steps from 15 participants
- Skill-level categorization including professional, expert, casual, and beginner drivers
- Trajectories covering at least five laps per track and car combination