新开源大规模导航基准

新开源大规模导航基准是由德克萨斯大学奥斯汀分校学习代理研究组创建，旨在为自主导航研究提供一个统一和全面的测试平台。该数据集包含500个高度约束的障碍课程环境，用于评估不同的学习方法。数据集的创建过程涉及使用Gazebo模拟器进行高保真物理模拟，并与ROS集成，以便于从模拟直接转移到物理机器人。该数据集的应用领域主要集中在自主机器人导航，旨在解决深度强化学习在导航中的不确定性、安全性、数据效率和环境泛化等问题。

A new open-source large-scale navigation benchmark was developed by the Learning Agents Research Group at The University of Texas at Austin, with the purpose of providing a unified and comprehensive testbed for autonomous navigation research. This dataset contains 500 highly constrained obstacle course environments for evaluating diverse learning methods. The creation process of this dataset involves using the Gazebo simulator for high-fidelity physical simulation and integrating it with ROS to enable direct transfer from simulated environments to physical robots. The primary application domain of this dataset focuses on autonomous robotic navigation, aiming to address core issues in deep reinforcement learning for navigation, including uncertainty, safety, data efficiency, and environmental generalization.