1.预测

1.1光栅化

代表性论文

1. Motion Prediction of Traffic Actors for Autonomous Driving using Deep Convolutional Networks (Uber)
2. MultiPath (Waymo)

问题

1. 渲染信息丢失
2. 感受野有限
3. 高计算复杂度

1.2图神经网络

1.2.1 图卷积

2. LaneGCN (uber 2020)

1.2.2 边卷积

1. VectorNet (waymo 2020)
   注意：Vectornet的子图使用的是边缘卷积，大图使用的是自注意力机制

Transformer

1. mmTransformer (2021)
2. WayFormer
3. QCNet (2023 CVPR，还没看懂)

1.3基于锚点

1. DenseTNT

1.4生成式模型

1. Social gan: Socially acceptable trajectories with generative adversarial networks (CVPR 2018)
2. Tranjectron++

1.5Metrics

1. b-minFDE：(1-p_i)^2 * minFDE_i
2. b-minADE：(1-p_i)^2 * minADE_i
   参考：https://eval.ai/web/challenges/challenge-page/454/evaluation

2.端到端

2.1光栅化

1. End-to-end Interpretable Neural Motion Planner (uber 2019)
2. P3 (uber)

2.2图神经网络

2.3Transformer

1. UniAD
2. VAD

2.4采样

1. Curve-based sampler (NMP )

2. Lane-based sampler (P3 uber 2020)

3. Retrieval-based sampler (MP3)

4. 如何理解agent centric