Sample Pipelines#
Sample pipelines are provided to demonstrate the capabilities of the Humanoid Toolkit. Review the validated hardware configuration before running a pipeline.
These pipelines are designed to showcase core Humanoid Toolkit features, including imitation learning, vision-based manipulation and SLAM. Each pipeline includes a detailed description, along with instructions for running the sample code on your system.
Train and evaluate an Action Chunking Transformer policy in simulation or on a robot.
Combine ACT, OCS2 model predictive control, and MuJoCo simulation.
Evaluate Transformer- and CNN-based diffusion policies on the Push-T task.
Run visual and visual-inertial SLAM with RGB-D, stereo, or monocular cameras.
Build and run point-cloud lidar-inertial odometry and mapping.
Use large-language-model planning and perception in a robotics workflow.
Deploy a diffusion-transformer policy for robot manipulation tasks.
Run a Pi0.5 vision-language-action model with real-time action chunking.
Deploy an OpenClaw agent with AgenticROS and an OpenVINO Model Server.
Run real-time lidar-inertial-visual odometry and mapping.
Build and run high-performance lidar-inertial odometry and mapping.
Explore whole-body control for GROOT-based humanoid robotics.
Optimize and deploy the GR00T N1.7 model with OpenVINO.