Learning to Drive (L2D)

Learning to Drive (L2D) 是由阿尔托大学计算机科学系创建的一个低成本真实世界强化学习基准。该数据集涉及一个简单可复制的实验设置，其中强化学习代理需要学习如何驾驶Donkey Car在三个微型赛道上行驶，仅使用单眼图像观测和车辆速度。数据集的创建过程包括详细的实验设置描述，确保了实验的可重复性。L2D数据集的应用领域主要集中在自主驾驶任务上，旨在通过强化学习算法解决从零开始学习驾驶的问题。

Learning to Drive (L2D) is a low-cost real-world reinforcement learning benchmark created by the Department of Computer Science at Aalto University. This dataset provides a simple and reproducible experimental setup, where reinforcement learning agents are required to learn to operate a Donkey Car on three miniature tracks using only monocular image observations and vehicle speed. The dataset creation process includes detailed descriptions of the experimental setup to ensure full experimental reproducibility. The L2D dataset is primarily focused on autonomous driving tasks, aiming to solve the problem of learning to drive from scratch via reinforcement learning algorithms.