Waymo Motion dataset
收藏资源简介:
该数据集基于Waymo Motion数据集,增加了详尽的帧级元动作标注,用于支持自动驾驶系统中的可控轨迹生成。数据集采用自回归元动作框架,将传统的长时间间隔元动作分解为帧级元动作,确保了轨迹段与其对应元动作之间的严格对齐,从而在整个轨迹跨度内实现了一致和统一的任务表述。数据集旨在解决现有框架中元动作与实际行为轨迹之间的时间错位问题,通过实验验证了该框架在轨迹适应性和动态决策场景中的响应性。
This dataset is developed based on the Waymo Motion dataset, with exhaustive frame-level meta-action annotations added to support controllable trajectory generation in autonomous driving systems. It adopts an autoregressive meta-action framework, which decomposes traditional long-interval meta-actions into frame-level meta-actions, ensuring strict alignment between trajectory segments and their corresponding meta-actions, thus achieving consistent and unified task representation across the entire trajectory span. This dataset aims to resolve the temporal misalignment between meta-actions and actual behavioral trajectories in existing frameworks, and experimental results have verified the framework’s trajectory adaptability and responsiveness in dynamic decision-making scenarios.




