Nocturne
收藏资源简介:
Nocturne是由Meta AI和UC Berkeley共同开发的2D驾驶模拟器数据集,专注于多智能体在部分可观测环境下的协调研究。该数据集包含134453个场景片段,利用开源轨迹和地图数据构建,支持从真实世界驾驶数据中加载和重放任意轨迹和场景。Nocturne旨在解决在无需计算机视觉和图像特征提取的计算开销下,研究真实世界多智能体环境中的推理和心智理论问题。数据集的应用领域包括强化学习和模仿学习,旨在提高智能体在复杂场景中的安全导航和协调能力。
Nocturne is a 2D driving simulator dataset co-developed by Meta AI and UC Berkeley, focusing on coordination research of multiple agents in partially observable environments. This dataset contains 134,453 scene segments, constructed using open-source trajectory and map data, and supports loading and replaying arbitrary trajectories and scenes from real-world driving data. Nocturne aims to enable research on reasoning and theory of mind in real-world multi-agent environments without the computational overhead imposed by computer vision and image feature extraction. Its application areas include reinforcement learning and imitation learning, with the goal of improving the safe navigation and coordination capabilities of agents in complex scenarios.

- 1Nocturne: a scalable driving benchmark for bringing multi-agent learning one step closer to the real worldMeta AI · 2023年



