behave: A behavior tree implementation in Python

With behave, you can define a behavior tree like this:

tree = (
    is_greater_than_10 >> wow_large_number
    | is_between_0_and_10 >> count_from_1
    | failer * repeat(3) * doomed
)

bb = tree.blackboard(10)
while bb.tick() == RUNNING:
    pass

Import from behave:

from behave import condition, action, FAILURE

Define condition nodes

Conditions are defined by functions return a bool value. The function can take any number of arguments.

@condition
def is_greater_than_10(x):
    return x > 10

@condition
def is_between_0_and_10(x):
    return 0 < x < 10

Define action nodes

Action functions can return SUCCESS, FAILURE, RUNNING states. Return None will be treated as SUCCESS.

@action
def wow_large_number(x):
    print "WOW, %d is a large number!" % x

@action
def doomed(x):
    print "%d is doomed" % x
    return FAILURE

And you can define an action with a generator function:

yield FAILURE fails the actions.
yield or yield None puts the action into RUNNING state.
If the generator returns(stop iteration), action state will be set to SUCCESS

@action
def count_from_1(x):
    for i in range(1, x):
        print "count", i
        yield
    print "count", x

Define a sequence

seq = is_greater_than_10 >> wow_large_number

Define a selector

sel = is_greater_than_10 | is_between_0_and_10

Decorate nodes

from behave import repeat, forever, succeeder

decorated_1 = forever(count_from_1)
decorated_2 = succeeder(doomed)
decorated_3 = repeat(10)(count_from_1)

For readability reason, you can also use chaining style:

from behave import repeat, forever, succeeder, failer

composite_decorator = repeat(3) * repeat(2)   # It's identical to repeat(6)

decorated_1 = forever * count_from_1
decorated_2 = succeeder * doomed
decorated_3 = repeat(10) * count_from_1
decorated_4 = failer * repeat(10) * count_from_1

Put everything together

tree = (
    is_greater_than_10 >> wow_large_number
    | is_between_0_and_10 >> count_from_1
    | failer * repeat(3) * doomed
)

Every node is a tree itself. And a big tree is composed by many sub trees. To iterate the tree:

bb = tree.blackboard(5) # Creates an run instance

# Now let the tree do its job, till job is done
state = bb.tick()
print "state = %s\n" % state
while state == RUNNING:
    state = bb.tick()
    print "state = %s\n" % state
assert state == SUCCESS or state == FAILURE

Output:

count 1
state = Running

count 2
state = Running

count 3
state = Running

count 4
state = Running

count 5
state = Success

Wait, did I mention debugger?

To debug the tree, you need to:

Define a debugger function
Create blackboard by calling tree.debug(debugger, arg1, arg2...) instead of tree.blackboard(arg1, arg2...).

def my_debugger(node, state):
    print "[%s] -> %s" % (node.name, state)

bb = tree.debug(my_debugger, 5) # Creates an blackboard with debugger enabled

# Now let the tree do its job, till job is done
state = bb.tick()
print "state = %s\n" % state
while state == RUNNING:
    state = bb.tick()
    print "state = %s\n" % state
assert state == SUCCESS or state == FAILURE

Output:

[ is_greater_than_10 ] -> Failure
[ BeSeqence ] -> Failure
[ is_between_0_and_10 ] -> Success
count 1
[ count_from_1 ] -> Running
[ BeSeqence ] -> Running
[ BeSelect ] -> Running
state = Running

count 2
[ count_from_1 ] -> Running
[ BeSeqence ] -> Running
[ BeSelect ] -> Running
state = Running

count 3
[ count_from_1 ] -> Running
[ BeSeqence ] -> Running
[ BeSelect ] -> Running
state = Running

count 4
[ count_from_1 ] -> Running
[ BeSeqence ] -> Running
[ BeSelect ] -> Running
state = Running

count 5
[ count_from_1 ] -> Success
[ BeSeqence ] -> Success
[ BeSelect ] -> Success
state = Success

Too messy? Let put some comments into the tree:

tree = (
    (is_greater_than_10 >> wow_large_number) // "if x > 10, wow"
    | (is_between_0_and_10 >> count_from_1) // "if 0 < x < 10, count from 1"
    | failer * repeat(3) * doomed // "if x <= 0, doomed X 3, and then fail"
)

And make a little change to my_debugger:

def my_debugger(node, state):
    if node.desc:
        print "[%s] -> %s" % (node.desc, state)

Try it again:

[ if x > 10, wow ] -> Failure
count 1
[ if 0 < x < 10, count from 1 ] -> Running
state = Running

count 2
[ if 0 < x < 10, count from 1 ] -> Running
state = Running

count 3
[ if 0 < x < 10, count from 1 ] -> Running
state = Running

count 4
[ if 0 < x < 10, count from 1 ] -> Running
state = Running

count 5
[ if 0 < x < 10, count from 1 ] -> Success
state = Success

Run Tests with nose

nosetests test

qinche7n/behave