Skip to content

README_MOTION_CONTROL.md

Latest commit

 

History

History
153 lines (118 loc) · 4.11 KB

README_MOTION_CONTROL.md

File metadata and controls

153 lines (118 loc) · 4.11 KB

Motion Control Demos for MuJoCo MCP

This directory contains motion control demonstrations showing how to control various robots through the MuJoCo MCP interface.

Available Demos

1. motion_control_demo.py - Direct Control Demo

A standalone demo that directly controls robots via the viewer server connection.

Features:

  • Load various robot models from MuJoCo Menagerie
  • Predefined motion patterns (wave, walk, grasp, etc.)
  • Interactive control mode
  • Support for multiple robot types:
    • Robotic Arms: Franka Panda, UR5e, Kuka iiwa
    • Quadrupeds: Anymal C, Unitree Go2, Spot
    • Humanoids: Unitree G1, H1
    • Grippers: Robotiq, Shadow Hand

Usage:

# Make sure viewer server is running
python mujoco_viewer_server.py

# In another terminal, run the demo
python examples/motion_control_demo.py

2. mcp_motion_control.py - MCP Interface Demo

Demonstrates control through the MCP protocol interface.

Features:

  • Natural language control
  • MCP tool integration
  • Demo sequences for basic scenes
  • Compatible with Claude Desktop

Usage:

# Run directly
python examples/mcp_motion_control.py

# Or use through Claude Desktop with MCP configured

Prerequisites

1. MuJoCo Menagerie

To use robot models from Menagerie:

# Clone MuJoCo Menagerie
git clone https://github.com/google-deepmind/mujoco_menagerie.git ~/mujoco_menagerie

# Or set environment variable to your installation
export MUJOCO_MENAGERIE_PATH=/path/to/mujoco_menagerie

2. Viewer Server

Always start the viewer server first:

python mujoco_viewer_server.py

Motion Control Examples

Basic Pendulum Control

# Create pendulum
await handle_call_tool("create_scene", {"scene_type": "pendulum"})

# Set position
await handle_call_tool("set_joint_positions", {"positions": [1.57]})  # 90 degrees

# Step simulation
await handle_call_tool("step_simulation", {"steps": 100})

Robot Arm Control

demo = MotionControlDemo()
demo.load_model("franka_panda")
demo.go_home()  # Move to home position
demo.wave_motion(duration=5.0)  # Wave for 5 seconds
demo.circle_motion(radius=0.2, duration=10.0)  # Circular motion

Natural Language Control

Through MCP interface:

  • "Create a pendulum simulation"
  • "Load Franka Panda robot"
  • "Move the robot arm in a circle"
  • "Make the robot wave"

Supported Robot Models

Model Type Joints Demo Motions
franka_panda Arm 7 wave, pick_place, circle
ur5e Arm 6 wave, pick_place, spiral
kuka_iiwa Arm 7 wave, figure8, reach
anymal_c Quadruped 12 stand, walk, trot
go2 Quadruped 12 stand, walk, jump
spot Quadruped 12 stand, walk, dance
g1 Humanoid 37 stand, wave_hand, walk
h1 Humanoid 25 stand, balance, squat
shadow_hand Hand 24 open, close, wave, grasp

Troubleshooting

  1. "Failed to connect to viewer server"

    • Make sure mujoco_viewer_server.py is running
    • Check if port 8888 is available

  2. "MuJoCo Menagerie not found"

    • Install MuJoCo Menagerie as shown above
    • Set MUJOCO_MENAGERIE_PATH environment variable

  3. "Model not found"

    • Check if the model exists in your Menagerie installation
    • Verify the path in robot_configs dictionary

Extending the Demos

To add new robot models:

  1. Add configuration to robot_configs in motion_control_demo.py:
"new_robot": {
    "path": "manufacturer_model/scene.xml",
    "type": "arm",  # or quadruped, humanoid, etc.
    "joints": 6,
    "home_position": [0, 0, 0, 0, 0, 0],
    "demo_motions": ["custom1", "custom2"]
}
  1. Implement custom motion patterns:
def custom_motion(self, duration: float = 5.0):
    """Custom motion pattern"""
    # Your motion control code here
  1. Add to demo_map in run_demo() method.

Advanced Features

  • Trajectory Planning: Implement smooth trajectories between positions
  • Force Control: Use torque control for compliant motions
  • Multi-Robot: Control multiple robots simultaneously
  • Sensor Feedback: Use state feedback for closed-loop control