Skip to main content

​
Overview

The height_control.py script provides a standalone simulation environment for the quadruped robot using the MuJoCo physics engine. You can run simulations with different terrain configurations and observe the robot’s gait controller in action.

​
Quick Start

1

Navigate to Source Directory

cd ~/workspace/source
2

Run Basic Simulation

Launch the simulation with rough terrain (default):
python3 height_control.py
The MuJoCo viewer window will open, showing the robot walking on rough terrain.
3

Control the View

The camera automatically follows the robot. You can interact with the viewer:
  • Rotate view: Click and drag with left mouse button
  • Pan: Click and drag with right mouse button
  • Zoom: Scroll mouse wheel
  • Pause/Resume: Press spacebar

​
Terrain Options

The simulation supports two terrain types:
Run with challenging terrain that includes obstacles and height variations:
python3 height_control.py --terrain rough
  • Model file: model/world_train.xml
  • Use case: Testing robustness and adaptive controllers
  • Expected behavior: Robot navigates uneven surfaces

​
Running Headless

For automated testing or data collection without the GUI, you can run simulations in headless mode by modifying the script to remove the viewer: 
# In height_control.py, replace the viewer loop with:
mujoco.mj_step(model, data)
# Process your data without visualization
Or create a headless wrapper script:
headless_sim.py
import mujoco
import numpy as np
from height_control import GAIT_PARAMS, apply_gait_targets
from gait_controller import DiagonalGaitController

model = mujoco.MjModel.from_xml_path("model/world_train.xml")
data = mujoco.MjData(model)
controller = DiagonalGaitController(GAIT_PARAMS)
controller.reset()

# Run for 1000 steps
for _ in range(1000):
    apply_gait_targets(controller, model.opt.timestep)
    mujoco.mj_step(model, data)
    # Collect data here

​
Gait Parameters

The default gait parameters are defined in height_control.py:17: 
GAIT_PARAMS = GaitParameters(
    body_height=0.05,     # 5cm body clearance
    step_length=0.06,     # 6cm forward stride
    step_height=0.04,     # 4cm foot lift
    cycle_time=0.8        # 0.8s per gait cycle
)

​
Expected Performance

# Computed from gait parameters
expected_velocity_x = step_length / cycle_time
# = 0.06m / 0.8s = 0.075 m/s

​
Troubleshooting

If you see warnings like:
[WARN] IK failed for leg FL with target [x, y, z]
This indicates the inverse kinematics solver couldn’t find a valid joint configuration. Common causes:
  • Target position is out of reach
  • Gait parameters are too aggressive
  • Solution: Reduce step_length or step_height in GAIT_PARAMS
If the robot becomes unstable:
  • Reduce step_height (default: 0.04)
  • Increase cycle_time (default: 0.8)
  • Lower body_height (default: 0.05)
  • Switch to flat terrain for testing
If the simulation runs slowly:
  • Close other applications
  • The simulation uses real-time stepping (see height_control.py:84)
  • Remove the sleep delay for faster-than-realtime execution

​
Key Source Files

  • height_control.py - Main simulation script
  • gait_controller.py - Diagonal gait pattern generator
  • ik.py - Inverse kinematics solver (3-DOF legs)
  • model/world.xml - Flat terrain definition
  • model/world_train.xml - Rough terrain definition

​
Next Steps

Train Adaptive Models

Learn to train PPO policies that adapt to terrain

ROS2 Integration

Connect the simulation to ROS2 for distributed control

Build docs developers (and LLMs) love

Get started for freeTalk to us