Skip to content

Install#

This page gives the requirements of ACRES and the procedure to get the code. The pages that follow build the game, ACRES Core and the ROS 2 workspace.

The Maxxum with a chisel plow on a field

Before You Start#

Make sure that the workstation has the items of this table.

Item Requirement Notes
Operating system Linux, 64 bit The test workstation uses Fedora 43.
Unreal Engine Version 5.8, source build Acres/Acres.uproject sets the engine association 5.8. The test workstation uses version 5.8.2.
Compiler The clang toolchain of the Unreal Engine source build The engine script Setup.sh installs it.
Git Git with Git LFS Git LFS holds the assets: 1537 files, 2.7 GB.
GPU Vulkan with shader model 6. Use a GPU with hardware ray tracing. The test workstation uses an NVIDIA GeForce RTX 5060 Ti with 16 GB. Without hardware ray tracing, the LiDAR uses its CPU ray caster.
Disk 20 GB free for the repository and the first build The packaged game uses 6.4 GB of this space. The engine source build needs more space.
conda Miniconda or an equivalent conda installation The Python tools and ROS 2 use conda environments.
ffmpeg On the PATH, with one of the encoders h264_nvenc, libx264, libopenh264 or mpeg4 Only the option -CaptureVideo= uses it.

The game selects the render tier from the memory of the GPU. A GPU with less than 11.5 GiB gets the low tier. A GPU with 11.5 GiB or more gets the high tier. Platforms and GPU Tiers gives the details.

Note

The packaged game does not need Unreal Engine, Git or conda. Copy the folder Packaged/Linux to a different workstation with a sufficient GPU and start Acres.sh.

Platforms#

The repository has settings for three platforms. Only Linux has a test record in this documentation.

Platform Graphics Status
Linux Vulkan, shader model 6 Built, packaged and tested. All commands of this documentation are Linux commands.
Windows DirectX 12, shader model 6 The Git history records a build, a cook and a package on Windows on 2026-09-30. Tools/Performance/benchmark_windows.ps1 measures that package. This documentation has no Windows procedure.
macOS Metal, shader model 6 The settings are in Acres/Config/DefaultEngine.ini. No build record exists.

Get the Code#

  1. Enable Git LFS one time for your user account.

    git lfs install

  2. Clone the repository. Replace <repository-url> with the address of the repository.

    git clone <repository-url> ACRES
cd ACRES

  3. Make sure that Git LFS downloaded the assets.

    ls -l Acres/Content/Maps/V03ACRE.umap

    Expected Result

    The size of the file is approximately 37 MB (37364256 bytes).

    CAUTION

    A size of some hundred bytes shows an LFS pointer file. The build fails with pointer files. Run git lfs pull and examine the size again.

Make the Conda Environments#

The repository uses three conda environments. Make only the environments that you need.

Environment Python Use Source
ros2 3.12 ROS 2 Humble from RoboStack: the ROS 2 bridge, the bags and the ROS 2 tools. ROS/Env/environment.yml
torchenv 3.13 The Python tools, ACRES Core, the package acres_learn and the PPO trainer. No file. See step 2.
acres-docs 3.12 The build of this documentation. Documentation/environment.yml

  1. Make the environment ros2 when you use ROS 2. Build the ROS 2 Workspace gives the full procedure.

    conda env create -f ROS/Env/environment.yml --override-channels -c robostack-humble -c conda-forge

  2. Make the environment torchenv when you use the Python tools or the learning stack.

    conda create -n torchenv python=3.13
conda activate torchenv
pip install -e Learning

    Note

    The repository has no environment file for torchenv. The script Learning/Scripts/ppo.sh activates an environment with this name. Learning/pyproject.toml lists the packages: numpy and matplotlib. The PPO trainer also needs torch and tensorboard. Build ACRES Core lists the packages of the Core build.

  3. Make the environment acres-docs when you build this documentation.

    conda env create -f Documentation/environment.yml

Next Steps#

Do the pages in this sequence.

  1. Build the Game: compile the game module and package the game.
  2. Build ACRES Core: build the simulator without Unreal Engine. The game does not need it.
  3. Build the ROS 2 Workspace: build the ROS 2 bridge. The game does not need it.
  4. Run the Packaged Game: start the game with the menu or with options.
  5. First Drive: drive the Maxxum.
  6. First ROS 2 Session: connect the game to ROS 2.
  7. First Recorded Dataset: record the camera and the LiDAR.