Back to the HomePage.

Quick start

Fast Downward is released in four flavours: tarball, Apptainer (formerly known as Singularity), Docker and Vagrant. Here we provide instructions to get you started as quickly as possible. You can find more usage information at PlannerUsage.

What flavour is for me?

• running experiments: We recommend Apptainer or the tarball. Docker is an alternative, but be mindful of its significant overhead.

• teaching: We recommend Vagrant.

• development: We recommend working on a clone of the master repository.

See WhatFlavourIsForMe for a more detailed discussion.

Running the Apptainer image

We have tested Apptainer 1.2.2. If Apptainer isn't installed on your machine, check the following section.

To download the Fast Downward image, run:

   1 apptainer pull fast-downward.sif docker://aibasel/downward:latest

Then run the planner. The example uses the LAMA-first configuration to solve a planning task located in the $BENCHMARKS directory. LAMA-first is designed to find solutions quickly without much regard for plan cost:

   1 ./fast-downward.sif --alias lama-first $BENCHMARKS/gripper/prob01.pddl

Unlike Docker (see below), $BENCHMARKS does not have to be an absolute path.

Installing Apptainer

Apptainer's predecessor, Singularity, used to be shipped with Ubuntu Linux for some time, making its installation very convenient. As of this writing, this is no longer the case, but Ubuntu (deb) packages are available from the Apptainer developers. For a typical Ubuntu system, download the AMD64 deb package from https://github.com/apptainer/apptainer/releases and install it like so (example for Apptainer 1.2.2):

   1 sudo apt install ./apptainer_1.2.2_amd64.deb

Running the Docker image

We assume that Docker is installed on your machine. You want to solve a planning problem located on the $BENCHMARKS directory. You can run the same LAMA-first configuration as before:

   1 sudo docker run --rm -v $BENCHMARKS:/benchmarks aibasel/downward --alias lama-first /benchmarks/gripper/prob01.pddl

Note the use of sudo (Docker usually requires root privileges).

Note that this mounts the local directory $BENCHMARKS of your host machine under the container directory /benchmarks, which is the place where the containerized planner looks for the problem. The path stored in the $BENCHMARKS variable must be absolute.

The Docker image for Fast Downward is installed on your machine as a side-effect of the command.

Using a Vagrant machine

We assume that

• Vagrant is installed on your machine

• The current directory contains the Fast Downward Vagrantfile for the desired release, the PDDL files domain.pddl and problem.pddl for the planning task you want to solve.

• The SoPlex LP solver is included automatically. If you want to also use the CPLEX LP solver within the planner, its installer file must be present in the directory /path/to/lp/installers. As of Fast Downward 23.06, you will need CPLEX 22.1.1 (installer filename cplex_studio2211.linux-x86-64.bin).

• You want to create your Vagrant VM as subdirectory my-fast-downward-vm of the current directory. The subdirectory does not exist yet.

Create and provision your virtual machine as follows:

   1 # Set up the VM. Only run this one.
   2 mkdir my-fast-downward-vm
   3 cp Vagrantfile my-fast-downward-vm/
   4 cp domain.pddl problem.pddl my-fast-downward-vm/
   5 # Skip next line if you don't need LP support.
   6 export DOWNWARD_LP_INSTALLERS=/path/to/lp/installers
   7 cd my-fast-downward-vm
   8 vagrant up
   9 # The VM is now set up.
  10 # You can now safely delete the LP installers and unset the environment variable.
  11 
  12 # Log into the VM and run the planner.
  13 vagrant ssh
  14 downward/fast-downward.py --alias lama-first /vagrant/domain.pddl /vagrant/problem.pddl
  15 
  16 # Log out from the VM.
  17 exit

Source code

See the build instructions for a complete description on how to build the planner from source. We recommend using the latest release, especially for scientific experiments. If you are using the main branch instead, be aware that things can break or degrade with every commit.

Next steps

• See other ways of invoking the planner on PlannerUsage.

• Read about recommended experiment setups.

• Go through the full Documentation.