This repository serves as the code hub for our entry into the multi agent programming contest. For this purpose we have created our own BDI module and communication protocol with the multi agent programming contest within Python.
pip install -r requirements
To use the code according to our implementation within the MAPC run python3 main.py.
The BDI-component as modelled in BDIAgent.py can be used independently. To use the BDI model separately create your agent class and have it be inherited like so:
from BDIAgent import BDIAgent
class SampleAgent(BDI):
...Within this SampleAgent there must be a run-function like so:
class SampleAgent(BDI):
def run():
while True:
new_intention = get_intention()
self.add_intention(new_intention)
action = self.execute_intention()
# perform actionThe get_intention()-function can be for example a series of if-statements that decides what intention the agent should perform next, but can be whatever you like as long as it returns an intention. Example:
class SampleAgent(BDI):
def run():
while True:
...
def get_intention():
if True:
return plan1
else:
return plan2The intention is then added to the intention queue through the BDI module. Next the intention is executed and an action is returned, this action should be a primitive function which the agent can directly execute (such as move). The new_intention is retrieved from what we call a plan within the SampleAgent-class. A plan can be created like this:
class SampleAgent(BDI):
def run():
...
def plan1():
# do some logic
intentions = [self.move, self.move]
args = [((2, 3),), ((1, 2),)]
contexts = [tuple(), tuple()]
descriptions = ["moveTo(2,3)", "moveTo(1,2)"]
primitives = [True, True]
return intentions, args, contexts, descriptions, primitivesA plan must always contain 5 lists, a list of intentions (which we model as a bound method), a list of arguments which are the arguments for the respective bound functions, contexts which are currently not used for anything but can be seen as what your agent should believe before executing the respective intention and primitives which is a list of Booleans that tell the model if a given intention is primitive or not. An intention is primitive if it can be directly executed by an agent, and it is non-primitive in case you want to add a sub-plan which the agent cannot directly perform. Example:
class SampleAgent(BDI):
def run():
while True:
new_intention = get_intention()
self.add_intention(new_intention)
action = self.execute_intention()
# perform action
def get_intention():
if True:
return plan1
else:
return plan2
def plan1():
# do some logic
intentions = [self.move, self.plan2]
args = [((2, 3),), tuple()]
contexts = [tuple(), tuple()]
descriptions = ["moveTo(2,3)", "plan2"]
primitives = [True, False]
return intentions, args, contexts, descriptions, primitives
def plan2():
# do some logic
intentions = [self.move]
args = [((1, 2),)]
contexts = [tuple()]
descriptions = ["moveTo(1,2)"]
primitives = [True]
return intentions, args, contexts, descriptions, primitivesThus the basic workflow of using the BDI model consists of having a run()-function, a get-intention()-function and creating plans and sub-plans.
- CCI Research group
- Mostafa Mohajeri Parizi
- Giovanni Sileno
- University of Amsterdam