ScenarioNet

Open-source platform for large-scale traffic scenario modeling and simulation

ScenarioNet

Open-source platform for large-scale traffic scenario modeling and simulation

Webpage | Code | Video | Paper | Browse Scenario

Meet ScenarioNet

ScenarioNet is an open-sourced platform for large-scale traffic scenario modeling and simulation.

  • ScenarioNet defines a unified scenario description format containing HD maps and detailed object annotations.
  • Scenarios recorded in this format can be replayed in the digital twins with multiple views, ranging from Bird-Eye-View layout to realistic 3D rendering.
  • ScenarioNet provides tools to build and manage databases built from various data sources including real-world datasets like Waymo, nuScenes, Lyft L5, and nuPlan datasets and synthetic datasets like the procedural generated ones and safety-critical ones.
  • We demonstrate several applications of ScenarioNet including large-scale scenario generation, AD testing, imitation learning, and reinforcement learning in both single-agent and multi-agent settings. The results imply scaling up the training data brings new research opportunities in machine learning and autonomous driving.

    • ScenarioNet's System Design

    Your Image
    ScenarioNet consists of the data layer, system layer, and application layer. Various datasets are unified into an internal scenario description. The system layer then provides a set of tools to operate on data efficiently, such as filtering, merging, sanity-check, splitting and so on. Once the database is ready, it can be loaded into MetaDrive for large-scale simulation and supports applications.

    Multi-sensory Simulation

    ScenarioNet leverages MetaDrive Simulator for multi-modal observation simulation.

    Traffic Scenarios from Various Datasets

    ScenarioNet reads from real world dataset such as Waymo, nuScenes, Lyft L5, and nuPlan datasets and creates interactive environment for closed-loop simulation.

    tSNE Visualization of Scenario Embeddings

    AD Stack Testing

    ScenarioNet bridges OpenPilot (Left) and ROS (Right) for autonomous driving testing.