gym-unrealcv Integration
This document covers the gym-unrealcv package for Reinforcement Learning research with UnrealCV environments.
Overview
The gym-unrealcv package (also called UnrealZoo) provides OpenAI Gym interfaces for UnrealCV-powered environments. It enables:
Reinforcement learning agent training
Embodied AI research
Navigation and tracking tasks
Multi-agent interactions
Repository: https://github.com/UnrealZoo/unrealzoo-gym
Installation
.. code-block:: bash
git clone https://github.com/UnrealZoo/unrealzoo-gym.git cd unrealzoo-gym pip install -e .
Dependencies:
unrealcv
gym
opencv-python
numpy
matplotlib
Quick Start
.. code-block:: python
import gymnasium as gym import gym_unrealcv
Create environment
env = gym.make(‘UnrealTrack-track_train-ContinuousColor-v0’)
Reset
obs, info = env.reset()
Step
action = env.action_space.sample() obs, reward, terminated, truncated, info = env.step(action)
Close
env.close()
Environment Naming
Format: Unreal{task}-{MapName}-{ActionSpace}{ObservationType}-v{version}
Tasks:
+—————————+—————————————-+
| Task | Description |
+—————————+—————————————-+
| Track | Object tracking |
| Navigation | Point navigation |
| Rendezvous | Target meeting |
+—————————+—————————————-+
Action Spaces:
+—————————+—————————————-+
| Space | Description |
+—————————+—————————————-+
| Discrete | Discretized actions |
| Continuous | Continuous control |
| Mixed | Mixed with interactive actions |
+—————————+—————————————-+
Observation Types:
+—————————+—————————————-+
| Type | Description |
+—————————+—————————————-+
| Color | RGB image |
| Depth | Depth map |
| Rgbd | RGB + depth |
| Gray | Grayscale |
| Mask | Segmentation mask |
| Pose | Agent pose |
+—————————+—————————————-+
Example: UnrealTrack-Greek_Island-ContinuousRgbd-v0
Available Maps
UE4 Examples:
track_trainGreek_IslandContainerYard_DayContainerYard_NightSuburbNeighborhood_DaySuburbNeighborhood_Night
UE5 Examples:
Map_ChemicalPlant_1Old_TownMiddleEastDemo_Roof
Documentation
Full documentation is available at:
Notion Page: https://unrealzoo.github.io/
Scene Gallery: https://www.notion.so/Scene-Gallery
API Docs: See gym_unrealcv/ directory
UnrealZoo Project
Paper: UnrealZoo: Enriching Photo-realistic Virtual Worlds for Embodied AI (ICCV 2025)
Key Features:
100+ photo-realistic scenes
Diverse entities (humans, vehicles, animals)
Multi-agent support (10+ agents)
Interactive actions (pick/drop objects)
Chaos system for vehicles
Citation
.. code-block:: bibtex
@misc{zhong2024unrealzooenrichingphotorealisticvirtual, title={UnrealZoo: Enriching Photo-realistic Virtual Worlds for Embodied AI}, author={Fangwei Zhong and Kui Wu and Churan Wang and Hao Chen and Hai Ci and Zhoujun Li and Yizhou Wang}, year={2024}, eprint={2412.20977}, archivePrefix={arXiv}, primaryClass={cs.AI}, url={https://arxiv.org/abs/2412.20977}, }