rofrano / kstar Goto Github PK

This can be run locally, in a docker container, or as a service.

## Suggested build for 64bit

./build.py release64

# ./fast-downward.py --build release64 <domain_file> <problem_file> --search "kstar(heuristic,k=<number-of-plans>)"

./fast-downward.py --build release64 examples/gripper/domain.pddl examples/gripper/prob01.pddl --search "kstar(blind(),k=100)"

• heurisitic: any heuristic provided by Fast Downward
  (http://www.fast-downward.org/Doc/Heuristic).
  Disclaimer: Optimality of K* is only guaranteed with an admissible and consistent heuristic.

You can run the Docker image locally by sharing your current folder and using a bash shell inside the container like this:

docker run --rm -it -v $(pwd):/work rofrano/kstar bash

/usr/bin/python3 /workspace/kstar/fast-downward.py --build release64 <domain_file> <problem_file> --search "kstar(heuristic,k=<number-of-plans>)"

For example (using files from git repo):

/usr/bin/python3 /workspace/kstar/fast-downward.py --build release64 examples/gripper/domain.pddl examples/gripper/prob01.pddl --search "kstar(blind(),k=100)"

Run the service locally via:

docker run -d -p 4501:4501 \
    --env server__hostname=`hostname`:4501 \
    --name <YOUR_CONTAINER_NAME> rofrano/kstar

A couple of notes:

• This will make the service available on your local machine on port 4501. See the Docker User Guide for manipulating port mappings and binding to specific interfaces.
• <YOUR_CONTAINER_NAME> can be any name like kstar or planner. It allows you to more easily manipulated it with commands like docker stop kstar, docker start kstar, and docker rm kstar
• The --env options are only necessary for the swagger UI available at (http://localhost:4501/api-docs) to properly bind to your physical host name and port instead of the container name. If you do not plan to use the swagger UI, you can safely skip it.

There is now a REST API service so that the planner can be called remotely. The REST API was intentionally kept consistent with the ctpelok77/ibmresearchaiplanningsolver API even though it only containes the kstar planner. It should allow easily swapping this planner out for the more robust version.

You can use the planner service with curl using the following commands:

url=http://localhost:4501/planners/topk/kstar-topk

domain=`sed 's/;/\n;/g' <DOMAIN-FILE> | sed '/^;/d' | tr -d '\n'`

problem=`sed 's/;/\n;/g' <PROBLEM-FILE> | sed '/^;/d' | tr -d '\n'`

body="{\"domain\": \"$domain\", \"problem\": \"$problem\", \"numplans\":<NUMBER-OF-PLANS>}"

basebody=`echo $body`

curl -d "$basebody" -H "Content-Type: application/json" "$url"

Where <DOMAIN-FILE> is your domain.pddl and <PROBLEM-FILE> is your problem.pddl and <NUMBER-OF-PLANS> is a positive integer that represents the number of plans you want to generate.

The response will be a json structure with all of the plans. Using the gripper example you would get something like this:

{
    "plans": [
        {
            "cost": 11,
            "actions": [
                "pick ball2 rooma left",
                "pick ball1 rooma right",
                "move rooma roomb",
                "drop ball1 roomb right",
                "drop ball2 roomb left",
                "move roomb rooma",
                "pick ball3 rooma left",
                "pick ball4 rooma right",
                "move rooma roomb",
                "drop ball3 roomb left",
                "drop ball4 roomb right"
            ]
        }
    ]
}

Michael Katz, Shirin Sohrabi, Octavian Udrea and Dominik Winterer
A Novel Iterative Approach to Top-k Planning [pdf] [bib]
In ICAPS 2018

For questions and comments please get in touch with Michael Katz ([email protected]).