ASPRILO, an intra-logistics benchmark suite for answer set programming https://asprilo.github.io/
see also github repository https://github.com/potassco/asprilo

This project aims to extend ASPRILO solver (using CLINGO) with MAPF family solvers, starting from M family problems (only moving toward shelves in the warehouse), future work would be to confront A B C family problems by reducing it to MAPF problem, and solving with MAPF solvers.

• While ASPRILO states sometimes being able to be installed on Windows platform, we have found it can be installed on Linux only (some commands exists only on Linux platform).
• Read carefully asprilo site and install the generator and vizualizer.
• Install dotnet core, in order to be able to run the MAPF solvers DLLs.
• Branches in current git:
  • master branch is handling Running Demo section
    create asprilo folder under users home directory, and clone this files and folders into it.
    master branch is following original ASPRILO installation while adding support for incremental solve,
    while working with vizualizer, see issue #31 on their git site: potassco/asprilo#31
    using this branch ASPRILO is solving only anonymous problems, while MAPF solvers are given default assignment
    (e.g. first robot to first shelf, second robot to second shelf... )
  • assignments branch is handling Running the tests section
    create asprilo folder under users home directory, and clone this files and folders into it.
    assignments branch has added support of assign rules to the encodings, so ASPRILO will solve non-anonymous problems

• update generate_problems.py file with your home directory
• run generate_problems.py
• problems will be generated into generatedInstances folder
• problems are spanning robots in the warehouse from 5 to 80 steping 5,
  • it is a medium size grid warehouse of 15X12
  • quantity of robots is equal to number of shelves to number of products to number of orders to number of total units.

• open asprilo_solver.py and update your home directory
• run asprilo_solver.py
  • it will run through all generated problems in generatedInstances folder
  • run ASPRILO solver against each problem and document solution and time (keep assignment of robots to shelves)
  • translate the prolbem into MAPF problem by generating agents file and map file
  • run dotnet solvers and document solutions steps and time solvers: A*+OD+ID, EPEA*, ICTS+ID, CBS, CBS modern (CBSH), MACBS over A*
  • create solvers_results.csv file with all documented results.

The ability of using the vizualizer to examine instances, and examine solvers solutions visually, has tremendous impact.
As this extended capability allowing the researcher easily compare and debug sovlers solutions.

open viz_demo.mp4 to see example functionality of the vizualizer working with the different solvers including(in the movie by order of appearence):

• ASPRILO solving anonymous and MAKESPAN goal
• A*+OD+ID solving non-anonymous (default assignment), goal MAKESPAN and Sum of Costs
• CBS solver exists only for MAKESPAN variant, so the solver solves when asking for MAKESPAN goal, and shows a message of non existent solver, when asking for Sum of Costs variant.

The vizualizer is communicating with the attached solver over configured port (5000 by default),
transmitting an instance problem, and expecting a plan solved by the solver, to show the solution in the vizualizer.
The workflow goes through this steps:

• initiate the solver that will work on the instance problem
• the solver opens a port listener and is waiting for incoming instance
• incoming instance is forwarded to initiated solver
  • with MAPF solvers, we changed the workflow to first convert ASP plan to MAPF format (map and agents files)
  • run a sub process triggering the .NET Dlls with the created files
  • on success, coverting back the solution to ASP format plan
• on solver success, plan is communicated back to the vizualizer for examining.

extended information on the changes is in SOLVER DEEP DIVE.md

• Eli Boyarski
• David Abekasis
• Nir Zagdanski

This project has no license

• We thank Dr.Roni Stern for the guidance and interesting lectures of Multi Agents Systems