Hierarchical Epistemic Multi-Agent Planner

Python library to encode in a declarative way the epistemic multi-agent planning problem.

How to install:

You can simply use pip install bash command:

pip install echosys

How to define planning objects:

After importing the library with:

from ECHO import *

you can define new types:

stack = IntType("stak", 1, 3)
color = EnumType("color", ["red", "orange", "yellow", "black"])
color_pair = StructType("colorxcolor", [color, color])

once the types are defined, you can define fluents:

on_block     = Fluent("on_block", color_pair)
top          = Fluent("top", color)
gripped      = Fluent("gripped", color)
free_gripper = Fluent("free_gripper")

note that free_gripper is to be considered as a boolean fluent.

Often it happens that you want to define ”schemata” actions, therefore, our library provides the means to define variables as well:

C1 = Variable("C1", color)
C2 = Variable("C2", color)
A1  = Variable("i", agent)

Classical actions are defined in the following way:

pick = IAction(name = "pick",
               params = [C1, C2],
               precondition = [top(C1), on_block(C1, C2), free_gripper()],
               effects = [-top(C1), -on_block(C1, C2), top(C2), -free_gripper(), gripped(C1)])

Note that free_gripper is a flunet, while free_gripper() is a literal. To negate a literal add - before the predicate.

Epistemic actions are defined in the following way:

pick_from_stack_place_on_table = MEAction('pspt',
  params = [A1, C1],
  precondition = [owner(A1, C1), free_table(), B([A1], free_table())],
  effects=[-owner(A1, C1), on_table(C1), -free_table()],
  full_obs=[A1])

The code is complient with the EPDDL syntax.

Basic types, fluents, variables and actions are then added to the planning problems, for instance:

p  = ClassicalPlanningProblem()
ep = MEPlanningProblem()

p.add_type(color)
p.add_fluent(free_table)
p.add_action(pick)

#and similar for ep

At the end you can combine the two planning problems with:

echo_problem = ECHOPlanningProblem(p, e)

How to solve the planning problem:

You can solve the following planning problems:

classical_plan = solve_classical(p)
epistemic_plan = solve_epddl(ep)
echo_plan = solve_echo(echo_problem)

To see a set of complete and working examples of planning problems defined and solved using ECHO python library we remind you to the examples folder.