English | 中文

An OpenWorldLib extension library built on the SimWorld simulation environment, providing core modules — Pipeline, Memory, Operator, and Representation — for multi-agent scenarios.

• SimWorldPipeline — Multi-agent simulation pipeline that drives agent behaviors step-by-step via policies
• SimWorld Representation — Wraps UnrealCV communication to manage pedestrians, vehicles, Robot Dogs, NPCs, and other scene objects
• Flexible Configuration — Customize simulation parameters via YAML files

• Python >= 3.10
• SimWorld UE client running (see launch instructions below)

Run the following script to automatically install Python dependencies and download the SimWorld UE client:

bash simworld_install.sh

The script auto-detects your OS (Linux / Windows via Git Bash) and downloads the corresponding SimWorld binary from Hugging Face.

To install only the Python package:

pip install -e .

Before running any tests or scripts, start the SimWorld UE client first.

Windows — Double-click SimWorld.exe, or launch from the command line:

./SimWorld.exe <MAP_PATH>

Linux — Run:

./SimWorld.sh <MAP_PATH>

For map configuration details, refer to the SimWorld Documentation.

from openworldlib_extensions.pipelines.simworld.pipeline_simworld import SimWorldPipeline

rep_cfg = dict(
    ip="127.0.0.1",
    port=9000,
    resolution=(640, 480),
    num_agents=3,
    spawn_position=[(0, 0), (200, 0), (400, 0)],
    # ... see src/.../data/default.yaml for all options
)

pipeline = SimWorldPipeline(rep_cfg=rep_cfg)
pipeline(policy)
pipeline.save_results(dir)

For a full working example, see test/test_simworld.py.

To customize the simulation scene (number of agents, spawn positions, map parameters, etc.), refer to:

src/openworldlib_extensions/representations/simulation_environment/simworld/simworld/data/default.yaml