def Match(team0, team1):
'''Run a single battle between two population files.'''
mod = module.getMod()
mod.load_team(team0, constants.OBJECT_TYPE_TEAM_0)
mod.load_team(team1, constants.OBJECT_TYPE_TEAM_1)
mod.set_speedup(100)
OpenNero.enable_ai()
def parseInputCommand(content):
"""
Parse commands from training window
"""
mod = getMod()
command, arg = content.attrib['command'], content.attrib['arg']
# first word is command rest is filename
if command.isupper():
vali = int(arg)
if command == "LT": mod.ltChange(vali)
if command == "EE": mod.eeChange(vali)
if command == "HP": mod.hpChange(vali)
if command == "SP": mod.set_speedup(vali)
if command == "save1": mod.save_team(arg, constants.OBJECT_TYPE_TEAM_0)
if command == "load1": mod.load_team(arg, constants.OBJECT_TYPE_TEAM_0)
if command == "rtneat": mod.deploy('rtneat')
if command == "qlearning": mod.deploy('qlearning')
if command == "pause": OpenNero.disable_ai()
if command == "resume": OpenNero.enable_ai()
if command == "example":
print 'command: example'
if arg == "start":
print 'command: example start'
mod.start_demonstration()
elif arg == "cancel":
print 'command: example cancel'
OpenNero.get_environment().cancel_demonstration()
elif arg == "confirm":
print 'command: example confirm'
OpenNero.get_environment().use_demonstration()
def parseInputCommand(content):
"""
Parse commands from training window
"""
mod = getMod()
command, arg = content.attrib['command'], content.attrib['arg']
# first word is command rest is filename
if command.isupper():
vali = int(arg)
if command == "LT": mod.ltChange(vali)
if command == "EE": mod.eeChange(vali)
if command == "HP": mod.hpChange(vali)
if command == "SP": mod.set_speedup(vali)
if command == "save1": mod.save_team(arg, constants.OBJECT_TYPE_TEAM_0)
if command == "load1": mod.load_team(arg, constants.OBJECT_TYPE_TEAM_0)
if command == "rtneat": mod.deploy('rtneat')
if command == "qlearning": mod.deploy('qlearning')
if command == "pause": OpenNero.disable_ai()
if command == "resume": OpenNero.enable_ai()
if command == "example":
print 'command: example'
if arg == "start":
print 'command: example start'
mod.start_demonstration()
elif arg == "cancel":
print 'command: example cancel'
OpenNero.get_environment().cancel_demonstration()
elif arg == "confirm":
print 'command: example confirm'
OpenNero.get_environment().use_demonstration()
def Match(team0, team1):
'''Run a single battle between two population files.'''
mod = module.getMod()
mod.load_team(team0, constants.OBJECT_TYPE_TEAM_0)
mod.load_team(team1, constants.OBJECT_TYPE_TEAM_1)
mod.set_speedup(100)
OpenNero.enable_ai()
def deploy(self, ai='rtneat', team=constants.OBJECT_TYPE_TEAM_0):
OpenNero.disable_ai()
if ai == 'rtneat':
OpenNero.set_ai('rtneat-%s' % team, None)
self.environment.remove_all_agents(team)
for _ in range(constants.pop_size):
self.spawnAgent(ai=ai, team=team)
OpenNero.enable_ai()
def deploy(self, ai='rtneat', team=constants.OBJECT_TYPE_TEAM_0):
OpenNero.disable_ai()
if ai == 'rtneat':
OpenNero.set_ai('rtneat-%s' % team, None)
self.environment.remove_all_agents(team)
for _ in range(constants.pop_size):
self.spawnAgent(ai=ai, team=team)
OpenNero.enable_ai()
def start_tower1(self): #Problem reduction
""" start the tower demo """
self.num_disks = 3
OpenNero.disable_ai()
self.stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
self.set_environment(env)
self.agent_id = common.addObject("data/shapes/character/BlocksRobot.xml", OpenNero.Vector3f(constants.GRID_DX, constants.GRID_DY, 2), type=constants.AGENT_MASK, scale=OpenNero.Vector3f(3,3,3))
OpenNero.enable_ai()
def start_my_planner_2_disk():
""" start the tower demo """
getMod().num_disks = 2
OpenNero.disable_ai()
getMod().stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
getMod().set_environment(env)
getMod().agent_id = common.addObject("data/shapes/character/MyPlanningRobot2.xml", OpenNero.Vector3f(TowerofHanoi.constants.GRID_DX, TowerofHanoi.constants.GRID_DY, 2), type=TowerofHanoi.constants.AGENT_MASK, scale=OpenNero.Vector3f(3,3,3))
OpenNero.enable_ai()
def start_nlp_extended(): #Natural Language Processing
""" start the tower demo """
getMod().num_disks = 3
OpenNero.disable_ai()
getMod().stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
getMod().set_environment(env)
getMod().agent_id = common.addObject("data/shapes/character/MyNLPRobot.xml", OpenNero.Vector3f(TowerofHanoi.constants.GRID_DX, TowerofHanoi.constants.GRID_DY, 2), type=TowerofHanoi.constants.AGENT_MASK, scale=OpenNero.Vector3f(3,3,3))
OpenNero.enable_ai()
def start_fps(self):
print 'start_fps was called'
if self.first_person_agent is None:
print 'adding first person agent!'
self.spawnAgent(agent_xml = 'data/shapes/character/FirstPersonAgent.xml')
OpenNero.enable_ai()
else:
print 'removing first person agent!'
common.removeObject(self.first_person_agent)
self.first_person_agent = None
def start_tower3(self): #2 Disk Goal Stack Planner
""" start the tower demo """
self.num_disks = 2
OpenNero.disable_ai()
self.stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
self.set_environment(env)
#self.set_environment(TowerEnvironment())
self.agent_id = common.addObject("data/shapes/character/BlocksRobot3.xml", OpenNero.Vector3f(constants.GRID_DX, constants.GRID_DY, 2), type=constants.AGENT_MASK, scale=OpenNero.Vector3f(3,3,3))
OpenNero.enable_ai()
def start_fps(self):
print 'start_fps was called'
if self.first_person_agent is None:
print 'adding first person agent!'
self.spawnAgent(
agent_xml='data/shapes/character/FirstPersonAgent.xml')
OpenNero.enable_ai()
else:
print 'removing first person agent!'
common.removeObject(self.first_person_agent)
self.first_person_agent = None
def add_bots(self, bot_type, num_bots):
OpenNero.disable_ai()
num_bots = int(num_bots)
if bot_type.lower().find("script") >= 0:
self.distribute_bots(num_bots, "data/shapes/roomba/Roomba.xml")
OpenNero.enable_ai()
return True
elif bot_type.lower().find("rtneat") >= 0:
self.start_rtneat(num_bots)
return True
else:
return False
def start_rtneat(self, pop_size):
" start the rtneat learning demo "
OpenNero.disable_ai()
#self.environment = RoombaEnvironment(constants.XDIM, constants.YDIM, self)
#set_environment(self.environment)
#self.reset_sandbox()
# Create RTNEAT object
rbound = OpenNero.FeatureVectorInfo()
rbound.add_continuous(-sys.float_info.max, sys.float_info.max)
rtneat = OpenNero.RTNEAT("data/ai/neat-params.dat", 2, 1, pop_size, 1.0, rbound, False)
rtneat.set_weight(0,1)
OpenNero.set_ai("rtneat",rtneat)
OpenNero.enable_ai()
self.distribute_bots(pop_size, "data/shapes/roomba/RoombaRTNEAT.xml")
def start_tower1(self): #Problem reduction
""" start the tower demo """
self.num_disks = 3
OpenNero.disable_ai()
self.stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
self.set_environment(env)
self.agent_id = common.addObject(
"data/shapes/character/BlocksRobot.xml",
OpenNero.Vector3f(constants.GRID_DX, constants.GRID_DY, 2),
type=constants.AGENT_MASK,
scale=OpenNero.Vector3f(3, 3, 3))
OpenNero.enable_ai()
def parseInputCommand(content):
"""
Parse commands from training window
"""
mod = getMod()
command, arg = content.attrib['command'], content.attrib['arg']
# first word is command rest is filename
if command.isupper():
vali = int(arg)
if command == "save1": mod.save_team(arg, constants.OBJECT_TYPE_TEAM_0)
if command == "load1": mod.load_team(arg, constants.OBJECT_TYPE_TEAM_0)
if command == "rtneat": mod.deploy('rtneat', 'neat')
if command == "qlearning": mod.deploy('none', 'qlearning')
if command == "pause": OpenNero.disable_ai()
if command == "resume": OpenNero.enable_ai()
def start_tower3(self): #2 Disk Goal Stack Planner
""" start the tower demo """
self.num_disks = 2
OpenNero.disable_ai()
self.stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
self.set_environment(env)
#self.set_environment(TowerEnvironment())
self.agent_id = common.addObject(
"data/shapes/character/BlocksRobot3.xml",
OpenNero.Vector3f(constants.GRID_DX, constants.GRID_DY, 2),
type=constants.AGENT_MASK,
scale=OpenNero.Vector3f(3, 3, 3))
OpenNero.enable_ai()
def parseInputCommand(content):
"""
Parse commands from training window
"""
mod = getMod()
command, arg = content.attrib['command'], content.attrib['arg']
# first word is command rest is filename
if command.isupper():
vali = int(arg)
if command == "save1": mod.save_team(arg, constants.OBJECT_TYPE_TEAM_0)
if command == "load1": mod.load_team(arg, constants.OBJECT_TYPE_TEAM_0)
if command == "rtneat": mod.deploy('rtneat', 'neat')
if command == "qlearning": mod.deploy('none', 'qlearning')
if command == "pause": OpenNero.disable_ai()
if command == "resume": OpenNero.enable_ai()
def start_demonstration(self):
'''
start the keyboard agent to collect demonstration example
'''
OpenNero.disable_ai()
team = constants.OBJECT_TYPE_TEAM_0
self.curr_team = team
#self.environment.remove_all_agents(team)
location = (self.spawn_x[team], self.spawn_y[team], 2)
agnt = common.addObject("data/shapes/character/steve_keyboard.xml",
position=OpenNero.Vector3f(*location),
type=team)
OpenNero.enable_ai()
self.environment.start_tracing()
return agnt
def start_agent_state_space_search(self): #State-space search
""" start the tower demo """
self.num_disks = 3
OpenNero.disable_ai()
self.stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
self.set_environment(env)
#self.set_environment(TowerEnvironment())
self.agent_id = common.addObject(
"data/shapes/character/BlocksRobot2.xml",
OpenNero.Vector3f(constants.GRID_DX, constants.GRID_DY, 2),
type=constants.AGENT_MASK,
scale=OpenNero.Vector3f(3, 3, 3))
OpenNero.enable_ai()
def start_nlp_extended(): #Natural Language Processing
""" start the tower demo """
getMod().num_disks = 3
OpenNero.disable_ai()
getMod().stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
getMod().set_environment(env)
getMod().agent_id = common.addObject(
"data/shapes/character/MyNLPRobot.xml",
OpenNero.Vector3f(TowerofHanoi.constants.GRID_DX,
TowerofHanoi.constants.GRID_DY, 2),
type=TowerofHanoi.constants.AGENT_MASK,
scale=OpenNero.Vector3f(3, 3, 3))
OpenNero.enable_ai()
def start_my_planner_3_disk():
""" start the tower demo """
getMod().num_disks = 3
OpenNero.disable_ai()
getMod().stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
getMod().set_environment(env)
getMod().agent_id = common.addObject(
"data/shapes/character/MyPlanningRobot3.xml",
OpenNero.Vector3f(TowerofHanoi.constants.GRID_DX,
TowerofHanoi.constants.GRID_DY, 2),
type=TowerofHanoi.constants.AGENT_MASK,
scale=OpenNero.Vector3f(3, 3, 3))
OpenNero.enable_ai()
def start_agent_state_space_search(self): # State-space search
""" start the tower demo """
self.num_disks = 3
OpenNero.disable_ai()
self.stop_agent()
env = TowerEnvironment()
env.initialize_blocks()
self.set_environment(env)
# self.set_environment(TowerEnvironment())
self.agent_id = common.addObject(
"data/shapes/character/BlocksRobot2.xml",
OpenNero.Vector3f(constants.GRID_DX, constants.GRID_DY, 2),
type=constants.AGENT_MASK,
scale=OpenNero.Vector3f(3, 3, 3),
)
OpenNero.enable_ai()
def start_demonstration(self):
'''
start the keyboard agent to collect demonstration example
'''
OpenNero.disable_ai()
team = constants.OBJECT_TYPE_TEAM_0
self.curr_team = team
#self.environment.remove_all_agents(team)
location = (self.spawn_x[team], self.spawn_y[team], 2)
agnt = common.addObject(
"data/shapes/character/steve_keyboard.xml",
position = OpenNero.Vector3f(*location),
type=team)
OpenNero.enable_ai()
self.environment.start_tracing()
return agnt
def add_bots(self, bot_type, num_bots):
OpenNero.disable_ai()
num_bots = int(num_bots)
if bot_type.lower().find("script") >= 0:
self.distribute_bots(num_bots, "data/shapes/roomba/Roomba.xml")
OpenNero.enable_ai()
return True
elif bot_type.lower().find("rtneat") >= 0:
self.start_rtneat(num_bots)
return True
elif bot_type.lower().find("rlagent") >= 0:
self.distribute_bots(num_bots, "data/shapes/roomba/RLAgent.xml")
OpenNero.enable_ai()
return True
else:
return False
def parseInput(strn):
if strn == "deploy" or len(strn) < 2:
return
mod = getMod()
# first word is command rest is filename
loc, val = strn.split(' ', 1)
vali = 1
if strn.isupper():
vali = int(val)
if loc == "HP": mod.hpChange(vali)
if loc == "SP": mod.set_speedup(vali)
if loc == "load1": mod.load_team(val, constants.OBJECT_TYPE_TEAM_0)
if loc == "load2": mod.load_team(val, constants.OBJECT_TYPE_TEAM_1)
if loc == "rtneat":
mod.deploy('rtneat', constants.OBJECT_TYPE_TEAM_0)
mod.deploy('rtneat', constants.OBJECT_TYPE_TEAM_1)
if loc == "qlearning":
mod.deploy('qlearning', constants.OBJECT_TYPE_TEAM_0)
mod.deploy('qlearning', constants.OBJECT_TYPE_TEAM_1)
if loc == "pause": OpenNero.disable_ai()
if loc == "resume": OpenNero.enable_ai()
def parseInput(strn):
if strn == "deploy" or len(strn) < 2:
return
mod = getMod()
# first word is command rest is filename
loc, val = strn.split(' ',1)
vali = 1
if strn.isupper():
vali = int(val)
if loc == "HP": mod.hpChange(vali)
if loc == "SP": mod.set_speedup(vali)
if loc == "load1": mod.load_team(val, constants.OBJECT_TYPE_TEAM_0)
if loc == "load2": mod.load_team(val, constants.OBJECT_TYPE_TEAM_1)
if loc == "rtneat":
mod.deploy('rtneat', constants.OBJECT_TYPE_TEAM_0)
mod.deploy('rtneat', constants.OBJECT_TYPE_TEAM_1)
if loc == "qlearning":
mod.deploy('qlearning', constants.OBJECT_TYPE_TEAM_0)
mod.deploy('qlearning', constants.OBJECT_TYPE_TEAM_1)
if loc == "pause": OpenNero.disable_ai()
if loc == "resume": OpenNero.enable_ai()
def start_rtneat(self):
""" start the rtneat learning stuff"""
self.spawnAgent()
OpenNero.enable_ai()
def start_rtneat(self):
""" start the rtneat learning stuff"""
self.spawnAgent()
OpenNero.enable_ai()
class NeroModule:
def __init__(self):
self.environment = None
self.agent_id = None
self.flag_loc = None
self.flag_id = None
self.set_speedup(constants.DEFAULT_SPEEDUP)
x = constants.XDIM / 2.0
y = constants.YDIM / 3.0
self.spawn_x = {}
self.spawn_y = {}
self.set_spawn(x, y, constants.OBJECT_TYPE_TEAM_0)
self.set_spawn(x, 2 * y, constants.OBJECT_TYPE_TEAM_1)
# Bounds for sensors in neural network and advice language. These bounds are
# used to convert sensor values between network and advice.
self.sbounds_network = OpenNero.FeatureVectorInfo()
self.sbounds_advice = OpenNero.FeatureVectorInfo()
# Networks have better learning bias when cube sensors produce values in the
# range [-1, 1], but the range [0, 1] is more intuitive in the advice
# language. Wall sensors use the same range [0, 1] for both network and advice.
# The sense() method in the ForageEnvironment class use these network bounds
# to scale the sensor values.
for i in range(constants.N_SENSORS):
self.sbounds_network.add_continuous(0, 1)
self.sbounds_advice.add_continuous(0, 1)
# The last sensor is the bias, which always takes the value 1 (upper bound).
self.sbounds_network.add_continuous(0, 1)
self.sbounds_advice.add_continuous(0, 1)
print 'sbounds_network', self.sbounds_network
def setup_map(self):
"""
setup the test environment
"""
OpenNero.disable_ai()
if self.environment:
error("Environment already created")
return
# create the environment - this also creates the rtNEAT object
self.environment = self.create_environment()
OpenNero.set_environment(self.environment)
# world walls
height = constants.HEIGHT + constants.OFFSET
common.addObject("data/shapes/cube/Cube.xml",
OpenNero.Vector3f(constants.XDIM / 2, 0, height),
OpenNero.Vector3f(0, 0, 90),
scale=OpenNero.Vector3f(constants.WIDTH,
constants.XDIM,
constants.HEIGHT * 2),
label="World Wall0",
type=constants.OBJECT_TYPE_OBSTACLE)
common.addObject("data/shapes/cube/Cube.xml",
OpenNero.Vector3f(0, constants.YDIM / 2, height),
OpenNero.Vector3f(0, 0, 0),
scale=OpenNero.Vector3f(constants.WIDTH,
constants.YDIM,
constants.HEIGHT * 2),
label="World Wall1",
type=constants.OBJECT_TYPE_OBSTACLE)
common.addObject("data/shapes/cube/Cube.xml",
OpenNero.Vector3f(constants.XDIM, constants.YDIM / 2,
height),
OpenNero.Vector3f(0, 0, 0),
scale=OpenNero.Vector3f(constants.WIDTH,
constants.YDIM,
constants.HEIGHT * 2),
label="World Wall2",
type=constants.OBJECT_TYPE_OBSTACLE)
common.addObject("data/shapes/cube/Cube.xml",
OpenNero.Vector3f(constants.XDIM / 2, constants.YDIM,
height),
OpenNero.Vector3f(0, 0, 90),
scale=OpenNero.Vector3f(constants.WIDTH,
constants.XDIM,
constants.HEIGHT * 2),
label="World Wall3",
type=constants.OBJECT_TYPE_OBSTACLE)
# Add an obstacle wall in the middle
common.addObject("data/shapes/cube/Cube.xml",
OpenNero.Vector3f(constants.XDIM / 2,
constants.YDIM / 2, height),
OpenNero.Vector3f(0, 0, 90),
scale=OpenNero.Vector3f(constants.WIDTH,
constants.YDIM / 4,
constants.HEIGHT * 2),
label="World Wall4",
type=constants.OBJECT_TYPE_OBSTACLE)
# Add some trees
for i in (0.25, 0.75):
for j in (0.25, 0.75):
# don't collide with trees - they are over 500 triangles each
common.addObject("data/shapes/tree/Tree.xml",
OpenNero.Vector3f(i * constants.XDIM,
j * constants.YDIM,
constants.OFFSET),
OpenNero.Vector3f(0, 0, 0),
scale=OpenNero.Vector3f(1, 1, 1),
label="Tree %d %d" % (10 * i, 10 * j),
type=constants.OBJECT_TYPE_LEVEL_GEOM)
# collide with their trunks represented with cubes instead
common.addObject(
"data/shapes/cube/Cube.xml",
OpenNero.Vector3f(i * constants.XDIM, j * constants.YDIM,
constants.OFFSET),
OpenNero.Vector3f(0, 0, 0),
scale=OpenNero.Vector3f(1, 1, constants.HEIGHT),
type=constants.OBJECT_TYPE_OBSTACLE)
# Add the surrounding Environment
common.addObject("data/terrain/NeroWorld.xml",
OpenNero.Vector3f(constants.XDIM / 2,
constants.YDIM / 2, 0),
scale=OpenNero.Vector3f(1.2, 1.2, 1.2),
label="NeroWorld",
type=constants.OBJECT_TYPE_LEVEL_GEOM)
return True
def create_environment(self):
return NeroEnvironment.NeroEnvironment()
def remove_flag(self):
if self.flag_id:
common.removeObject(self.flag_id)
def change_flag(self, loc):
if self.flag_id:
common.removeObject(self.flag_id)
self.flag_loc = OpenNero.Vector3f(*loc)
self.flag_id = common.addObject("data/shapes/cube/BlueCube.xml",
self.flag_loc,
label="Flag",
scale=OpenNero.Vector3f(1, 1, 10),
type=constants.OBJECT_TYPE_FLAG)
def place_basic_turret(self, loc):
return common.addObject("data/shapes/character/steve_basic_turret.xml",
OpenNero.Vector3f(*loc),
type=constants.OBJECT_TYPE_TEAM_1)
#The following is run when one of the Deploy buttons is pressed
def deploy(self, ai='rtneat', team=constants.OBJECT_TYPE_TEAM_0):
OpenNero.disable_ai()
if ai == 'rtneat':
OpenNero.set_ai('rtneat-%s' % team, None)
self.environment.remove_all_agents(team)
for _ in range(constants.pop_size):
self.spawnAgent(ai=ai, team=team)
OpenNero.enable_ai()
#The following is run when the Save button is pressed
def save_team(self, location, team=constants.OBJECT_TYPE_TEAM_0):
# if there are rtneat agents in the environment, save them as a group.
rtneat = OpenNero.get_ai("rtneat-%s" % team)
if rtneat:
location = rtneat.save_population(str(location))
# then, check whether there are any qlearning agents, and save them.
with open(location, 'a') as handle:
for agent in self.environment.teams[team]:
if agent.group == 'Agent' and agent.ai == 'qlearning':
handle.write('\n\n%s' % agent.to_string())
if hasattr(agent, 'stats'):
handle.write('\n\n%s' % agent.stats())
#The following is run when the Load button is pressed
def load_team(self, location, team=constants.OBJECT_TYPE_TEAM_0):
OpenNero.disable_ai()
self.environment.remove_all_agents(team)
if not os.path.exists(location):
print location, 'does not exist, cannot load population'
return
# parse out different agents from the population file.
contents = ''
try:
try:
handle = gzip.open(location)
contents = handle.read()
finally:
handle.close()
except Exception, e:
with open(location) as handle:
contents = handle.read()
if not contents:
print 'cannot read', location, 'skipping'
return
rtneat, qlearning, stats = self._split_population(
contents.splitlines(True))
print 'qlearning agents:', qlearning.count('Approximator')
# load any qlearning agents first, subtracting them from the population
# size that rtneat will need to manage. since we cannot deserialize an
# agent's state until after it's been added to the world, we put the
# serialized chunk for the agent into a map, then NeroEnvironment#step
# takes care of the deserialization.
pop_size = constants.pop_size
if qlearning.strip():
for chunk in re.split(r'\n\n+', qlearning):
if not chunk.strip():
continue
id = self.spawnAgent(ai='qlearning', team=team)
self.environment.agents_to_load[id] = chunk
pop_size -= 1
if pop_size == 0:
break
print 'rtneat agents:', rtneat.count('genomeend')
# load any rtneat agents from the file, as a group.
if pop_size > 0 and rtneat.strip():
tf = tempfile.NamedTemporaryFile(delete=False)
tf.write(rtneat)
tf.close()
print tf.name
OpenNero.set_ai(
"rtneat-%s" % team,
OpenNero.RTNEAT(tf.name, "data/ai/neat-params.dat", pop_size,
rtneat_rewards(), False))
os.unlink(tf.name)
while pop_size > 0:
self.spawnAgent(ai='rtneat', team=team)
pop_size -= 1
OpenNero.enable_ai()
