The Unified Planning Service

Offers unified-planning (version 1.2.0) as a service to the AI on Demand platform through Protobuf/gRPC.

• Protobuf Messages and Service Definition

How to Run

Locally

In a Python 3.9 environment, run:

pip install -r requirements.txt
python run.py

Docker Compose

docker-compose build
docker-compose up

AI4Experiments

Usage

The following example can be found ./examples/test_client_one_shot.py.

Create a Planning Problem

Create a problem with the unified-planning API. The following is the basic example used in the unified-planning repository README.

    x = Fluent("x")

    a = InstantaneousAction("a")
    a.add_precondition(Not(x))
    a.add_effect(x, True)

    problem = Problem("basic")
    problem.add_fluent(x)
    problem.add_action(a)
    problem.set_initial_value(x, False)
    problem.add_goal(x)

Connect to a Unified Planning Server

Create a channel with host name and port and use the channel to create a stub:

    planner_channel = grpc.insecure_channel('0.0.0.0:8061')
    planner_stub = up_pb2_grpc.UnifiedPlanningStub(planner_channel)

Create a Protobuf Request

Using the ProtoBuf writer supplied by the unfified-planning, we can create a Protobuf version of our problem and wrap it into a planning request:

    from unified_planning.grpc.proto_writer import ProtobufWriter
    writer = ProtobufWriter()
    
    pb_problem = writer.convert(problem)

    request = up_pb2.PlanRequest(
        problem=pb_problem,
        resolution_mode=up_pb2.PlanRequest.SATISFIABLE,
    )

Call the Unified Planning Service

Finally, we can call the unified-planning service and print the result:

    result = planner_stub.planOneShot(request)
    reader = ProtobufReader()
    result_up = reader.convert(result, problem)
    print(result_up.plan)

Known Issues / TODO

• Add remaining planning modes
• Install additional engines in container
• Compilation service does not include compilation type information
• Server-side PlanAnytime stream not stopped when client stops