Practical overview
When it fits
Use Puppeteer when you want to study or experiment with hierarchical humanoid control in the repository’s MuJoCo and DMControl tasks, rather than a general-purpose AI model or physical-robot deployment package.
Who it is for
- Robotics and machine-learning researchers studying humanoid control.
- Developers experimenting with simulated movement and visual terrain navigation.
Documented uses
- Evaluate released low- and high-level model checkpoints.
- Train models for tracking, walking or corridor navigation.
- Compare simulated locomotion with vision-based navigation across terrain obstacles.
Requirements & limits
The public GitHub repository is licensed under MIT; its README notes that third-party code has separate licenses. Model checkpoints are linked from the repository.
- Training requires a GPU with at least 24 GB of memory.
- The documented environment options are Docker and Conda.
- The repository provides eight tasks: stand, walk, run, corridor, hurdles-corridor, gaps-corridor, walls-corridor and stairs-corridor.
- The README recommends default hyperparameters and suggests Weights & Biases for tracking training.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Evaluate a released corridor-navigation checkpoint
Evaluate a downloaded high-level checkpoint for a visual corridor task using its low-level tracking checkpoint.
Before you start
- A Docker or Conda environment with the repository dependencies.
- A machine with a GPU containing at least 24 GB of memory for training; the repository states this requirement for training, not evaluation.
- Downloaded low-level tracking and high-level corridor checkpoints.
Steps
Set up the dependencies using the provided Dockerfile or the Conda environment file.
Download the low-level tracking and high-level corridor checkpoints.
From the repository, run the command below, replacing the checkpoint paths with the downloaded files.
python evaluate.py task=corridor low_level_fp=/path/to/tracking.pt checkpoint=/path/to/corridor-1.pt
Expected outcome
Runs the documented evaluation workflow for the corridor task with a low-level tracking checkpoint and a high-level corridor checkpoint.
Guide limitations
- The guide follows the repository's example command; it does not establish that a particular setup or run has been tested.
- The README says high-level checkpoints were trained with the same low-level checkpoint.
Set up Puppeteer’s dependencies
Prepare the repository environment for the documented training and inference workflows.
Before you start
- A machine with Docker, or Conda and the system packages recommended in the repository Dockerfile.
Steps
For Docker setup, enter the repository's
dockerdirectory.Build the image with the command below.
docker build . -t /puppeteer:1.0.0Alternatively, create the Conda environment with
conda env create -f docker/environment.yaml.
Expected outcome
Creates the provided Docker image or Conda environment for the repository's training and inference dependencies.
Guide limitations
- Manual Conda setup may require additional dependencies depending on existing system packages; consult the repository Dockerfile for recommended system packages.
- This is a dependency setup guide; it does not cover running a training or evaluation task.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- nicklashansen/puppeteer · Official homepage title · retained in its source language