OpenCDA-MARL Cooperative Intersection Benchmark Dataset
Description
This dataset packages the uploaded assets used by the OpenCDA-MARL framework for cooperative autonomous intersection management. It includes two CARLA intersection map archives, traffic recording/configuration files for multiple arrival-rate settings, and trained model checkpoints for released MARL baselines.
Included Assets
- 2 CARLA map packages for the 12-lane intersection benchmark
- 7 JSON recording/configuration files spanning lite and 100-500 VPH traffic settings
- Best checkpoints for DQN, MAPPO, SAC, and TD3 benchmark runs
Related Project and Paper
This dataset accompanies the OpenCDA-MARL project and the paper OpenCDA-MARL: A Unified Benchmarking Framework for Cooperative Autonomous Intersection Management With Multi-Agent Reinforcement Learning. Use the repository for framework code, environment setup, and experiment configuration details.
Recommended Uses
- Benchmarking cooperative multi-agent reinforcement learning policies at intersections
- Reproducing traffic-demand experiments across multiple vehicle-per-hour settings
- Evaluating pretrained policies and comparing checkpoint behavior across algorithms
- Studying map variants and traffic recordings used in the OpenCDA-MARL framework
Technical Details
Tags
Dataset Files
Main Dataset File
2 files
• 487.1 MB
Expand files
Base OpenCDA-MARL intersection map package for cooperative intersection experiments.
Alternate map package with the BEV-fixed straight-through traffic variant.
Sample Data
7 files
• 0.74 MB
Expand files
Traffic recording/configuration JSON used for scenario generation and replay.
Traffic recording/configuration JSON used for scenario generation and replay.
Traffic recording/configuration JSON used for scenario generation and replay.
Traffic recording/configuration JSON used for scenario generation and replay.
Traffic recording/configuration JSON used for scenario generation and replay.
Traffic recording/configuration JSON used for scenario generation and replay.
Traffic recording/configuration JSON used for scenario generation and replay.
Code/Scripts
8 files
• 223.2 MB
Expand files
Best trained model weights for this OpenCDA-MARL benchmark configuration.
Best trained model weights for this OpenCDA-MARL benchmark configuration.
Best trained model weights for this OpenCDA-MARL benchmark configuration.
Best trained model weights for this OpenCDA-MARL benchmark configuration.
Best trained model weights for this OpenCDA-MARL benchmark configuration.
Best trained model weights for this OpenCDA-MARL benchmark configuration.
Best trained model weights for this OpenCDA-MARL benchmark configuration.
Best trained model weights for this OpenCDA-MARL benchmark configuration.
Citation
Guo, L., and collaborators. OpenCDA-MARL Cooperative Intersection Benchmark Dataset. Associated with OpenCDA-MARL: A Unified Benchmarking Framework for Cooperative Autonomous Intersection Management With Multi-Agent Reinforcement Learning. University of Arizona Radar Lab.
Download Dataset
Authentication Required
Create a free account or sign in to download this dataset and its files.
Already have an account? Sign in above.