f1tenth_gym

该数据集是一个二维赛车模拟器，其中包含了一个带有侧滑的单轨自行车模型，并使用了光线行进技术进行二维激光雷达模拟。该模拟器支持离线优化代理，以及对抗性环境下的反事实遗憾预测训练。其规模可容纳多个代理在赛车场景中进行互动，任务旨在测试在竞争性策略和高动作规划中的对抗性设置。

This dataset is a 2D racing simulator that incorporates a single-track bicycle model with sideslip, and employs the ray marching technique for 2D LiDAR simulation. This simulator supports offline agent optimization, as well as counterfactual regret prediction training in adversarial environments. It has scalable capacity to accommodate multiple agents interacting in racing scenarios, with its core tasks designed to test adversarial settings involving competitive strategies and high-complexity motion planning.