ScenarioNet
Open-Source Platform for Large-Scale Traffic Scenario Simulation and Modeling
ScenarioNet is an open-source platform developed by MetaDriverse for large-scale traffic scenario modeling and simulation. It provides tools to build and manage databases from various data sources, including real-world datasets like Waymo Open Dataset, nuScenes, Lyft Level 5, and nuPlan. ScenarioNet defines a unified scenario description format containing HD maps and detailed object annotations, enabling scenarios to be replayed in digital twins with multiple views, ranging from bird's-eye view layouts to realistic 3D rendering. arXiv+8metadriverse.github.io+8NeurIPS+8
The platform consists of three layers:
Data Layer: Unifies various datasets into an internal scenario description.metadrive-simulator.readthedocs.io+5metadriverse.github.io+5NeurIPS+5
System Layer: Provides tools to operate on data efficiently, such as filtering, merging, sanity checks, and splitting.metadriverse.github.io
Application Layer: Enables loading the database into MetaDrive for large-scale simulation and supports applications like autonomous driving testing, imitation learning, and reinforcement learning in both single-agent and multi-agent settings. Google Colab+8metadriverse.github.io+8arXiv+8 ScenarioNet leverages the MetaDrive Simulator for multi-modal observation simulation and bridges platforms like OpenPilot and ROS for autonomous driving testing. It supports scenarios from various datasets and creates interactive environments for closed-loop simulation. metadriverse.github.io+1GitHub+1
For more information, you can visit the official ScenarioNet webpage or explore the GitHub repository.