ABOUT

This is a real-time solver framework. We implement algorithms such as LSSLRTA*, Nancy, Data-Driven Nancy.

Paper

[1] Maximilian Fickert, Tianyi Gu, Leonhard Staut, Wheeler Ruml, Joerg Hoffmann, and Marek Petrik, Beliefs We Can Believe In: Replacing Assumptions with Data in Real-Time Search. Proceedings of the Thirty-fourth AAAI Conference on Artificial Intelligence (AAAI-20), 2020.

[pdf] [slides] [poster]

[2] Maximilian Fickert, Tianyi Gu, Leonhard Staut, Sai Lekyang, Wheeler Ruml, Joerg Hoffmann, and Marek Petrik, Real-time Planning as Data-driven Decision-making. Proceedings of the ICAPS Workshop on Bridging the Gap Between AI Planning and Reinforcement Learning (PRL-20), 2020.

[pdf] [talk] [slides] [poster]

How to build

make

How to run

Single Run (solve one problem instance)

Usage:
  ./realtimeSolver [OPTION...] < [input file]

 -d, --domain arg          domain type: randomtree, tile, pancake, racetrack
                                (default: racetrack)
 -s, --subdomain arg       puzzle type: uniform, inverse, heavy, sqrt;
                           pancake type: regular, heavy;racetrack map :
                           barto-bigger, hanse-bigger-double, uniform (default:
                           barto-bigger)
 -a, --alg arg             realtime algorithm: bfs, astar, fhat,
                           lsslrtastar, risk, riskdd, riskddSquish (default: risk)
 -l, --lookahead arg       expansion limit (default: 100)
 -o, --performenceOut arg  performence Out file (default: out.txt)
 -v, --pathOut arg         path Out file
 -h, --help                Print usage

Run batch of experiments

cd scipts/testHarnesses

./multiThread-realtimeSolver.sh 
[-f instance]                    default: 1
[-n # of instances to test]      default: 100
[-d domain]                      default: pancake
[-s subdomain]                   default: regular
[-z domain size]                 default: 16
[-a algorithm ]
 support list,eg: -a a1 -a a2    default: risk, riskddSquish, lsslrtastar
[-l lookahead, ]
 support list,eg: -l 10 -l 30    default: 3, 10, 30, 100, 300, 1000
[-e algorithm extention          default: null]
[-t thread number]               default: 1
[-h help]

Training Repo

Source code of offline training is avaiable here

Problem Instances

All the instances can be found here