Skip to content

Installation

This section describes how to install and set up the PACE (Precise Adaptation through Continuous Evolution) framework on your system. PACE is built on top of NVIDIA Isaac Lab and requires a working GPU-accelerated simulation environment.

System Requirements

Supported Operating Systems

  • Ubuntu 20.04 or newer (tested and recommended: Ubuntu 22.04)

Hardware

  • NVIDIA GPU with CUDA support (RTX series recommended)
  • At least 8 GB GPU memory (16 GB recommended for large-scale experiments)
  • 16 GB system RAM (32 GB recommended)

Software

  • NVIDIA Driver
  • CUDA Toolkit (compatible with your Isaac Sim version)
  • Python >= 3.10
  • Git

PACE is developed and tested with Isaac Sim 5.0, Isaac Lab 0.46.2 and Python 3.11.13. Other versions may work but are not officially supported at the moment.

Isaac Sim compatibility notes

PACE is designed around the physics capabilities introduced in Isaac Sim 5.0, including: - Joint viscous friction modeling - Improved articulation parameter handling - More consistent actuator behavior

When running PACE on Isaac Sim versions < 5.0, you may see warnings such as:

Setting joint viscous friction coefficients are not supported in Isaac Sim < 5.0

This means the simulator silently ignores these parameters. As a result: - The simulation will run, but - The actuator dynamics will not match the intended PACE model - Sim-to-real performance may degrade

For accurate system identification and controller transfer, we strongly recommend using:

Isaac Sim 5.0 + Isaac Lab (current stable)

Future versions of PACE may rely even more heavily on features introduced in Isaac Sim 5.x.

1. Install Isaac Lab

PACE is implemented as an extension for Isaac Lab. Please follow the official Isaac Lab installation instructions first:

👉 https://isaac-sim.github.io/IsaacLab/

2. Clone the PACE repository

Clone the PACE repository (outside of IsaacLab):

cd ~
git clone https://github.com/leggedrobotics/pace-sim2real.git

3. Install the PACE extension

Activate your Conda environment and install PACE

cd ~/pace-sim2real
# use 'PATH_TO_isaaclab.sh|bat -p' instead of 'python' if Isaac Lab is not installed in Python venv or conda
python -m pip install -e source/pace_sim2real

All required Python dependencies (including CMA-ES) are installed automatically during this step.

You are now ready to run PACE 🚀

Proceed to:

  • Examples – run your first experiment
  • Guids – how to set up your own robot and tasks
  • Concepts – core ideas behind PACE and sim-to-real transfer