Lyft Level 5 Prediction

他的数据集包括我们的自动车队遇到的汽车，骑自行车的人，行人和其他交通代理的运动日志。这些日志来自通过我们团队的感知系统处理原始激光雷达、摄像机和雷达数据，非常适合训练运动预测模型。该数据集由捕获自动驾驶车辆周围环境的170,000场景组成。每个场景都会编码给定时间点车辆周围的状态。
该基线解决方案对来自包含在数据集内的代理位置的200万多个样本进行训练。该模型一次预测单个代理。首先，栅格生成一个鸟瞰图 (BEV) 自顶向下的栅格，该栅格对所有代理和地图进行编码。网络根据此栅格推断代理的未来坐标。

This dataset includes motion logs of cars, cyclists, pedestrians, and other traffic agents encountered by our autonomous fleet. These logs are generated by processing raw LiDAR, camera, and radar data via our team’s perception system, making it highly suitable for training motion prediction models. This dataset comprises 170,000 scenarios that capture the surrounding environment of autonomous driving vehicles, with each scenario encoding the surrounding state of the host vehicle at a given timestamp.

The baseline solution is trained on over 2 million samples derived from the agent positions contained within the dataset. This model predicts a single agent at a time. First, a top-down bird's-eye view (BEV) grid is generated, which encodes all traffic agents and the surrounding map. The network then infers the future coordinates of the target agent based on this grid.

搜集汇总

背景与挑战

背景概述

Lyft Level 5 Prediction是一个专注于自动驾驶运动预测的数据集，包含170,000个场景，这些场景基于Lyft自动驾驶车队的激光雷达、摄像机和雷达数据生成，编码了车辆周围交通代理的状态。该数据集适用于训练运动预测模型，其基线方法通过鸟瞰图栅格处理超过200万个样本，以预测单个代理的未来轨迹。