CMP: Cooperative Motion Prediction with Multi-Agent Communication

Paper | Project page | arXiv

This repository contains the official implementation of paper "CMP: Cooperative Motion Prediction with Multi-Agent Communication".

Features

1) Practical, Latency-robust Framework for Cooperative Motion Prediction: our framwork integrates cooperative perception with trajectory prediction, marking a pioneering effort in the realm of connected and automated vehicles, which enables CAVs to share and fuse data from LiDAR point clouds to improve object detection, tracking, and motion prediction.

2) Attention-based Prediction Aggregation: prediction aggregator take advantage of the predictions shared by other CAVs, which improves prediction accuracy. This mechanism is scalable and can effectively handle varying numbers of CAVs.

3) State-of-the-art Performance in cooperative prediction under practical settings on the OPV2V and V2V4Real datasets: our framwork evaluated on both simulated V2V datasets and real world V2V scenarios, and outperforms the cooperative perception and prediction network proposed by the strongest baseline V2VNet.

Getting Started

• Environment Setup
• Prepare Dataset and Checkpoints
• Multi-Ego Perception Eval
• Multi-Ego Prediciton Eval
• Visualization

Visualizations

The visualizations of predicted trajectories (colored waypoints) and ground truth (black lines) under different model settings in different traffic scenarios.

CMP

No Cooperation

Citation

@misc{wang2024cmpcooperativemotionprediction,
      title={CMP: Cooperative Motion Prediction with Multi-Agent Communication}, 
      author={Zehao Wang and Yuping Wang and Zhuoyuan Wu and Hengbo Ma and Zhaowei Li and Hang Qiu and Jiachen Li},
      year={2024},
      eprint={2403.17916},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2403.17916}, 
}