examples

Examples

This directory contains example scripts for training policies in K-Sim.

While not strictly necessary, we recommend using ksim with a GPU - we do most of our development on 4090s, but other GPUs should work as well, although you might encounter minor bugs.

To get started, follow these instructions:

1. Clone this repository:

git clone git@github.com:kscalelabs/ksim.git
cd ksim

2. Create a new Python environment (requires Python 3.11 or later). We recommend using uv.

3. Install ksim:

pip install ksim  # To install the public version
pip install -e '.'  # To install your local copy

4. Run an example script:

python -m examples.walking

Notes

If you're getting segfaults on Mujoco on a Linux-based headless GPU machine, you can force CPU rendering:

export MUJOCO_GL=osmesa

This will require:

sudo apt-get install -y libosmesa6 libosmesa6-dev