def getMovement(layoutText, agentIndex, difficulty):
gameState = GameState()
numGhosts = layoutText.count('G')
layoutText = layoutText.split('\n')
layoutText = filter(None,layoutText)
layout = Layout(layoutText)
gameState.initialize(layout,numGhosts);
if difficulty == 1:
agent = ReflexAgent()
agent.index = agentIndex
return agent.getAction(gameState)
elif difficulty == 2:
agent = MinimaxAgent()
agent.index = agentIndex
return agent.getAction(gameState)
elif difficulty == 2:
agent = AlphaBetaAgent()
agent.index = agentIndex
agent.depth = 4
return agent.getAction(gameState)
else:
agent = RandomAgent()
agent.index = agentIndex
return agent.getAction(gameState)
def __init__(self,
ghostAgents,
display,
rules,
layout,
percentRandomize=0.5):
Game.__init__(self, ghostAgents, display, rules)
initState = GameState()
initState.initialize(layout, len(ghostAgents), percentRandomize)
self.state = initState
self.initialState = initState.deepCopy()
class MultiAgentSearchAgent(Agent):
"""
This class provides some common elements to all of your
multi-agent searchers. Any methods defined here will be available
to the MinimaxPacmanAgent, AlphaBetaPacmanAgent & ExpectimaxPacmanAgent.
You *do not* need to make any changes here, but you can if you want to
add functionality to all your adversarial search agents. Please do not
remove anything, however.
Note: this is an abstract class: one that should not be instantiated. It's
only partially specified, and designed to be extended. Agent (game.py)
is another abstract class.
"""
moveToAction = { Directions.WEST: PacmanAction.WEST,
Directions.EAST: PacmanAction.EAST,
Directions.NORTH: PacmanAction.NORTH,
Directions.SOUTH: PacmanAction.SOUTH,
Directions.STOP: PacmanAction.STOP,}
actionToMovement = {PacmanAction.WEST: Directions.WEST,
PacmanAction.EAST: Directions.EAST,
PacmanAction.NORTH: Directions.NORTH,
PacmanAction.SOUTH: Directions.SOUTH,
PacmanAction.STOP: Directions.STOP}
def __init__(self, evalFn = 'scoreEvaluationFunction', depth = '2'):
self.index = 0 # Pacman is always agent index 0
self.evaluationFunction = scoreEvaluationFunction
self.depth = int(depth)
self.gameState = GameState()
self.initializeGameState()
rospy.Subscriber("/pacman_interface/agent_action", AgentAction, self.agentActionCallback)
rospy.Service('get_action', PacmanGetAction, self.getAction)
rospy.spin()
def initializeGameState(self):
rospy.wait_for_service('pacman_inialize_game_state')
try:
getInitializationInfo = rospy.ServiceProxy('pacman_inialize_game_state', PacmanInitializationInfo)
initInfo = getInitializationInfo()
thisLayout = layout.getLayout(initInfo.layout)
numGhosts = initInfo.numGhosts
self.gameState.initialize(thisLayout, numGhosts)
print "Game initialized"
except rospy.ServiceException, e:
print "Service call failed: %s"%e
def newGame(self,
layout,
pacmanAgent,
ghostAgents,
display,
quiet=False,
catchExceptions=False):
from pacman import GameState
agents = [pacmanAgent] + ghostAgents[:layout.getNumGhosts()]
initState = GameState()
initState.initialize(layout, len(ghostAgents))
game = GameExtended(agents,
display,
self,
catchExceptions=catchExceptions)
game.state = initState
self.initialState = initState.deepCopy()
self.quiet = quiet
return game
def getMovement(layoutText, agentIndex, difficulty):
gameState = GameState()
numGhosts = layoutText.count('G')
layoutText = layoutText.split('\n')
layoutText = filter(None, layoutText)
layout = Layout(layoutText)
gameState.initialize(layout, numGhosts)
if difficulty == 1:
agent = ReflexAgent()
agent.index = agentIndex
direction = agent.getAction(gameState)
elif difficulty == 2:
agent = MinimaxAgent()
agent.index = agentIndex
direction = agent.getAction(gameState)
elif difficulty == 3:
agent = AlphaBetaAgent()
agent.index = agentIndex
agent.depth = 3
direction = agent.getAction(gameState)
else:
agent = RandomAgent()
agent.index = agentIndex
direction = agent.getAction(gameState)
if direction == 'North':
return 1
elif direction == 'South':
return 2
elif direction == 'West':
return 3
else:
return 4
% % %% %%% % % % % %%% %% % %
% % % %
% %%%%% %%%%%%% %%%%%%% %%%%% %
% %% % % % % % % %% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
"""
numGhosts = layoutText.count('G')
layoutText = layoutText.split('\n')
layoutText = filter(None,layoutText)
layout = Layout(layoutText)
x.initialize(layout,numGhosts);
agent = MinimaxAgent()
agent.index=1
randomAgent = RandomAgent()
randomAgent.index = 1
reflexAgent = ReflexAgent()
reflexAgent.index = 1
minimaxAgent = MinimaxAgent()
minimaxAgent.depth = 2
minimaxAgent.index = 1
alphabetaAgent = AlphaBetaAgent()
def generateGameState(args):
layout = generateLayout(args)
gameState = GameState()
numGhostAgents = 1
gameState.initialize(layout, numGhostAgents)
return gameState
def generateAllStates(length, ghostNum = 1): #length of all possible spaces. Do not set the ghost num
allStatesWithoutP = []
for k in range(0, 4**length):
layout = util.base10toN(k, 4, length)
allStatesWithoutP.append(layout)
allValidStates = []
for k in allStatesWithoutP:
zerocount = 0
for x in range(0, len(k)):
if k[x] == "0":
zerocount += 1
if zerocount == (ghostNum+1):
allValidStates.append(k)
allLayouts = []
for k in allValidStates: #hardcoded for only ONE GHOST!!
tempstring1 = ""
tempstring2 = ""
switcher = True
for x in range(0, len(k)):
if k[x] == "0":
if switcher:
tempstring1 += "4"
tempstring2 += "5"
else:
tempstring1 += "5"
tempstring2 += "4"
switcher = False
else:
tempstring1 += k[x]
tempstring2 += k[x]
allLayouts.append(tempstring1)
allLayouts.append(tempstring2)
for k in range(0, len(allLayouts)):
state = allLayouts[k]
newstate = "%"
for x in range(0, len(state)):
if state[x] == "1":
newstate+=" "
elif state[x] == "2":
newstate += "."
elif state[x] == "3":
newstate+= "o"
elif state[x] == "4":
newstate+= "P"
elif state[x] == "5":
newstate+= "G"
newstate+= "%"
layouttext = []
layouttext.append("%"*(length+2)) #HARDCODE
layouttext.append(newstate)
layouttext.append("%"*(length+2)) #HARDCODE
allLayouts[k] = layouttext
#print layouttext
allStates = []
for k in range(0, len(allLayouts)):
layout = Layout(allLayouts[k])
gameState = GameState()
gameState.initialize(layout, 1) #ghost hardcoded
allStates.append(gameState)
return allStates
def generateGameState(args):
layout = generateLayout(args)
gameState = GameState()
numGhostAgents = 1
gameState.initialize(layout, numGhostAgents)
return gameState
def generateAllStates(
length,
ghostNum=1): #length of all possible spaces. Do not set the ghost num
allStatesWithoutP = []
for k in range(0, 4**length):
layout = util.base10toN(k, 4, length)
allStatesWithoutP.append(layout)
allValidStates = []
for k in allStatesWithoutP:
zerocount = 0
for x in range(0, len(k)):
if k[x] == "0":
zerocount += 1
if zerocount == (ghostNum + 1):
allValidStates.append(k)
allLayouts = []
for k in allValidStates: #hardcoded for only ONE GHOST!!
tempstring1 = ""
tempstring2 = ""
switcher = True
for x in range(0, len(k)):
if k[x] == "0":
if switcher:
tempstring1 += "4"
tempstring2 += "5"
else:
tempstring1 += "5"
tempstring2 += "4"
switcher = False
else:
tempstring1 += k[x]
tempstring2 += k[x]
allLayouts.append(tempstring1)
allLayouts.append(tempstring2)
for k in range(0, len(allLayouts)):
state = allLayouts[k]
newstate = "%"
for x in range(0, len(state)):
if state[x] == "1":
newstate += " "
elif state[x] == "2":
newstate += "."
elif state[x] == "3":
newstate += "o"
elif state[x] == "4":
newstate += "P"
elif state[x] == "5":
newstate += "G"
newstate += "%"
layouttext = []
layouttext.append("%" * (length + 2)) #HARDCODE
layouttext.append(newstate)
layouttext.append("%" * (length + 2)) #HARDCODE
allLayouts[k] = layouttext
#print layouttext
allStates = []
for k in range(0, len(allLayouts)):
layout = Layout(allLayouts[k])
gameState = GameState()
gameState.initialize(layout, 1) #ghost hardcoded
allStates.append(gameState)
return allStates
class PacmanTask(Task):
def __init__(self, layout, agents, display, state_repr='stack',
muteAgents=False, catchExceptions=False):
'''
state_repr: state representation, possible values ['stack', 'k-frames', 'dict']
'stack' - stack walls, food, ghost and pacman representation into a 4D tensor.
'dict' - return the raw dict representation keys=['walls', 'food', 'ghost', 'pacman'], values are matrix/tensor.
'k-frames' - instead of directional descriptors for pacman and ghost, use static descriptors and capture past k frames.
'''
# parse state representation.
self.state_repr = state_repr
self.layout = layout
self.agents = agents
self.display = display
self.muteAgents = muteAgents
self.catchExceptions = catchExceptions
if self.state_repr.endswith('frames'):
bar_pos = self.state_repr.rfind('frames')
self.state_k = int(self.state_repr[:bar_pos-1])
self.state_history = []
self.init_state = GameState()
self.init_state.initialize(layout, len(agents))
self.game_rule = ClassicGameRules(timeout=100)
self.myagent = GameNoWaitAgent()
self.init_game = Game([self.myagent] + agents[:layout.getNumGhosts()],
display,
self.game_rule,
catchExceptions = catchExceptions)
self.init_game.state = self.init_state
# action mapping.
self.all_actions = Actions._directions.keys()
self.action_to_dir = {action_i: action
for (action_i, action) in enumerate(self.all_actions)}
self.dir_to_action = {action: action_i
for (action_i, action) in enumerate(self.all_actions)}
def start_game():
self.game = self.init_game.deepCopy()
self.game.init()
self.start_game = start_game
start_game()
def deep_copy(self):
agents = list(self.agents)
task = PacmanTask(self.layout, agents, self.display, self.state_repr, self.muteAgents, self.catchExceptions)
task.game = self.game.deepCopy()
task.myagent = self.myagent # TODO: agents not deep copy.
return task
@property
def curr_state_dict(self):
return self.game.state.data.array()
@property
def curr_state(self):
if self.state_repr == 'dict':
return self.curr_state_dict
elif self.state_repr == 'stack':
state_dict = self.curr_state_dict
if len(state_dict.get('ghosts')) == 0:
ghost_stacked = np.zeros_like(state_dict['pacman'])
else:
ghost_stacked = np.sum(state_dict['ghosts'], axis=0)
state = np.array(
[
state_dict['food'],
state_dict['wall']
]
+
[
state_dict['pacman'][:, :, i] for i in range(4)
]
+
[
ghost_stacked[:, :, i] for i in range(4)
]
)
return state
elif hasattr(self, 'state_k'):
state_history = self.state_history + [self.curr_state_dict]
def stack_direction(state_agent):
return np.sum(state_agent, axis=2)
state = []
k = 0
for hist_dict in state_history[::-1]:
state.extend([
hist_dict['food'],
hist_dict['wall'],
stack_direction(hist_dict['pacman'])
]
+
sum(
[
[stack_direction(ghost) for ghost in hist_dict['ghosts']]
], []
)
)
k += 1
state = np.array(state)
frame_dim = state.shape[0] / k
for ki in range(k, self.state_k + 1):
state = np.concatenate((state, np.zeros_like(state[:frame_dim, :, :])), axis=0)
return state
def is_end(self):
return self.game.gameOver
@property
def num_actions(self):
return len(self.all_actions)
@property
def valid_actions(self):
dirs = self.game.state.getLegalPacmanActions()
return [self.dir_to_action[dir] for dir in dirs]
def step(self, action):
if hasattr(self, 'state_k'): # if we use past frames.
self.state_history.append(self.curr_state_dict)
if len(self.state_history) > self.state_k:
self.state_history = self.state_history[-self.state_k:]
if action not in self.valid_actions: # TODO: hack.
action = self.dir_to_action[Directions.STOP]
# convert action to direction.
direction = self.action_to_dir[action]
old_score = self.game.state.data.score
# run the game using the direction.
self.myagent.next_action = direction
self.game.run_one()
new_score = self.game.state.data.score
reward = new_score - old_score
if self.is_end():
self.game.finalize()
return reward
def reset(self):
self.start_game()
@property
def state_shape(self):
return self.curr_state.shape
def __str__(self):
return str(self.game.state)
def __repr__(self):
return str(self.game.state)
