Infinite synthetic data generation for embodied AI
luck-e.mp4
Hyperrealistic robotics simulation framework with Python API for embodied AI training and testing.
-
Download Lucky World Executable from our releases page and add its path to your system variables
# Set environment variables (choose one method): # Method 1: Set LUCKYWORLD_PATH directly to the executable export LUCKYWORLD_PATH=/path/to/LuckyWorld.exe # Windows export LUCKYWORLD_PATH=/path/to/LuckyWorld # Linux/Mac # Method 2: Set LUCKYWORLD_HOME to the directory containing the executable export LUCKYWORLD_HOME=/path/to/luckyworld/directory
-
Create conda environment (recommended)
conda create -n luckyrobots python conda activate luckyrobots
-
Install
pip install luckyrobots
-
Run Example
git clone https://github.com/luckyrobots/luckyrobots.git cd luckyrobots/examples python controller.py
from luckyrobots import LuckyRobots, Node
import numpy as np
# Create controller node
class RobotController(Node):
async def control_loop(self):
# Reset environment
reset_response = await self.reset_client.call(Reset.Request())
# Send actions
actuator_values = np.array([0.1, 0.2, -0.1, 0.0, 0.5, 1.0])
step_response = await self.step_client.call(Step.Request(actuator_values=actuator_values))
# Access observations
observation = step_response.observation
joint_states = observation.observation_state
camera_data = observation.observation_cameras
# Start simulation
luckyrobots = LuckyRobots()
controller = RobotController()
luckyrobots.register_node(controller)
luckyrobots.start(scene="kitchen", robot="so100", task="pickandplace")- so100: 6-DOF manipulator with gripper
- stretch_v1: Mobile manipulator
- dji300: Quadcopter drone
- kitchen: Residential kitchen environment
- loft: Open floor plan apartment
- drone_flight: Outdoor flight area
- pickandplace: Object manipulation
- navigation: Path planning and movement
LuckyRobots: Main simulation manager
start(scene, robot, task, observation_type): Initialize simulationregister_node(node): Add controller nodespin(): Run main loop
Node: Base class for robot controllers
create_client(service_type, service_name): Create service clientcreate_service(service_type, service_name, handler): Create service server
Reset: Reset robot to initial state
request = Reset.Request(seed=42, options={})
response = await reset_client.call(request)Step: Send action and get observation
request = Step.Request(actuator_values=[0.1, 0.2, -0.1])
response = await step_client.call(request)Access sensor data from step responses:
# Joint positions and velocities
joint_states = response.observation.observation_state
# Camera images (RGB + depth)
for camera in response.observation.observation_cameras:
image = camera.image_data # numpy array
name = camera.camera_name # "head_cam", "hand_cam", etc.# Basic usage
python controller.py --robot so100 --scene kitchen --task pickandplace
# With camera display
python controller.py --show-camera --rate 30
# Custom host/port
python controller.py --host 192.168.1.100 --port 3001Robot configurations are defined in src/luckyrobots/config/robots.yaml:
so100:
action_space:
actuator_names: [shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper]
actuator_limits:
- name: shoulder_pan
lower: -2.2
upper: 2.2
available_scenes: [kitchen]
available_tasks: [pickandplace]Lucky Robots uses a distributed node architecture:
- Manager Node: Central message routing
- LuckyRobots Node: Simulation interface
- Controller Nodes: User-defined robot controllers
- WebSocket Transport: Inter-node communication
- Lucky World: Physics simulation backend
git clone https://github.com/luckyrobots/luckyrobots.git
cd luckyrobots
pip install -e .- Fork the repository
- Create a feature branch
- Make changes and add tests
- Submit a pull request
MIT License - see LICENSE file.
- Issues: GitHub Issues
- Discussions: GitHub Discussions
- Discord: Community Server