Online Mapping and Topology Reasoning

1. LaneGraph2Seq: Lane Topology Extraction with Language Model via Vertex-Edge Encoding and Connectivity Enhancement [arXiv][Github]

• Using Language model for output decoding.
• Pros: Better connectivity and consistency
• Cons: Discretization leads to low accuracy

2. LANESEGNET: MAP LEARNING WITH LANE SEGMENT PERCEPTION FOR AUTONOMOUS DRIVING [arXiv][Github]

• Joint learning of lane boundaries and centerline formulation.

3. Translating Images to Road Network: A Non-Autoregressive Sequence-to-Sequence Approach [arXiv]
4. TopoMLP
5. TopoNet
6. MapTRv2
7. MapTR
8. STSU
9. VectorMapNet
10. HDMapNet

Leveraging SD maps

1. Neural Map Prior for Autonomous Driving arXiv

Downstream tasks for Joint Learning

1. VAD

Building Blocks

View Transform Module

1. LSS
2. BEVFormer
3. BEVFusion

Decoder

1. Deformable DETR
2. Language Models

Representation Learning

1. V-JEPA [Paper]
2. CLIP [arXiv][Blog]

Scene Reconstruction

1. REAL-TIME PHOTOREALISTIC DYNAMIC SCENE REPRESENTATION AND RENDERING WITH 4D GAUSSIAN SPLATTING [arXiv][Github]

Labs

1. Marco Pavone
2. Hang Zhao
3. Xingang Wang
4. Li Zhang