import os
import random
import sys
import time
import re
import uuid
from collections import namedtuple
from operator import add
EntityInfo = namedtuple(
'EntityInfo',
'x, y, z, yaw, pitch, name, colour, variation, quantity, life')
EntityInfo.__new__.__defaults__ = (0, 0, 0, 0, 0, "", "", "", 1, "")
# Create one agent host for parsing:
agent_hosts = [MalmoPython.AgentHost()]
# Parse the command-line options:
agent_hosts[0].addOptionalFlag("debug,d", "Display debug information.")
agent_hosts[0].addOptionalIntArgument(
"agents,n", "Number of agents to use, including observer.", 4)
try:
agent_hosts[0].parse(sys.argv)
except RuntimeError as e:
print('ERROR:', e)
print(agent_hosts[0].getUsage())
exit(1)
if agent_hosts[0].receivedArgument("help"):
print(agent_hosts[0].getUsage())
exit(0)
print "Bailing now."
exit(1)
time.sleep(0.1)
print ".",
if time.time() - start_time >= time_out:
print "Timed out while waiting for mission to start - bailing."
exit(1)
print
print "Mission has started."
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w',
0) # flush print output immediately
# -- set up two agent hosts --
agent_host_simeon = MalmoPython.AgentHost()
agent_host_fred = MalmoPython.AgentHost()
try:
agent_host_simeon.parse(sys.argv)
except RuntimeError as e:
print 'ERROR:', e
print agent_host_simeon.getUsage()
exit(1)
if agent_host_simeon.receivedArgument("help"):
print agent_host_simeon.getUsage()
exit(0)
# -- set up the mission --
xml = '''<?xml version="1.0" encoding="UTF-8" standalone="no" ?>
<Mission xmlns="http://ProjectMalmo.microsoft.com" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<Name>MalmoTutorialBot</Name>
<AgentStart>
<Placement x="0.5" y="56.0" z="0.5"/>
</AgentStart>
<AgentHandlers>
<AgentQuitFromTouchingBlockType>
<Block type="redstone_block"/>
</AgentQuitFromTouchingBlockType>
</AgentHandlers>
</AgentSection>
</Mission>'''
# Create default Malmo objects:
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse( sys.argv )
except RuntimeError as e:
print('ERROR:',e)
print(agent_host.getUsage())
exit(1)
if agent_host.receivedArgument("help"):
print(agent_host.getUsage())
exit(0)
if agent_host.receivedArgument("test"):
num_repeats = 1
else:
num_repeats = 10
def start_default_mission(agent_host):
my_mission = MalmoPython.MissionSpec()
my_mission_record = MalmoPython.MissionRecordSpec()
start_mission(agent_host, my_mission, my_mission_record)
def main():
sight = {'x': (-30, 30), 'z': (-30, 30), 'y': (-1, 1)}
range_x = abs(sight['x'][1] - sight['x'][0]) + 1
range_y = abs(sight['y'][1] - sight['y'][0]) + 1
range_z = abs(sight['z'][1] - sight['z'][0]) + 1
malmoutils.fix_print()
agent_host = MalmoPython.AgentHost()
malmoutils.parse_command_line(agent_host)
recordingsDirectory = malmoutils.get_recordings_directory(agent_host)
recordingsDirectory = "../human_trajectories"
if (not os.path.exists(recordingsDirectory)):
os.mkdir(recordingsDirectory)
logging.basicConfig(level=logging.INFO)
# pdb.set_trace()
logger = logging.getLogger(__name__)
logger.setLevel(logging.DEBUG) # set to INFO if you want fewer messages
video_width = 640
video_height = 480
sys.argv
mission_xml_path = "../custom_xmls/usar.xml"
validate = True
# my_mission = MalmoPython.MissionSpec(missionXML, validate)
my_mission = MalmoPython.MissionSpec(getMissionXML(mission_xml_path),
validate)
# ObservationFromGrid
my_mission.observeGrid(sight['x'][0], sight['y'][0], sight['z'][0],
sight['x'][1], sight['y'][1], sight['z'][1],
'relative_view')
# agent_host.setObservationsPolicy(MalmoPython.ObservationsPolicy.LATEST_OBSERVATION_ONLY)
agent_host.setVideoPolicy(MalmoPython.VideoPolicy.LATEST_FRAME_ONLY)
if agent_host.receivedArgument("test"):
num_reps = 1
else:
num_reps = 30000
my_mission_record = MalmoPython.MissionRecordSpec()
if recordingsDirectory:
my_mission_record.recordRewards()
my_mission_record.recordObservations()
my_mission_record.recordCommands()
# if agent_host.receivedArgument("record_video"): # my_mission_record.recordMP4(24,2000000)
my_mission_record.recordMP4(24, 2000000)
recording_name = datetime.now().strftime("%Y-%m-%d_%I-%M-%S_%p")
for iRepeat in range(1):
my_mission_record.setDestination(
os.path.join(recordingsDirectory, recording_name + ".tgz"))
max_retries = 3
for retry in range(max_retries):
try:
agent_host.startMission(my_mission, my_mission_record)
break
except RuntimeError as e:
if retry == max_retries - 1:
logger.error("Error starting mission: %s" % e)
exit(1)
else:
time.sleep(2)
logger.info('Mission %s', iRepeat)
logger.info("Waiting for the mission to start")
world_state = agent_host.getWorldState()
while not world_state.has_mission_begun:
print(".", end="")
time.sleep(0.1)
world_state = agent_host.getWorldState()
print()
img_counter = 0
# print('observations', world_state.observations)
while world_state.is_mission_running:
world_state = agent_host.getWorldState()
# Observations
# msg = observe(agent_host)
# if msg is not None:
# print('timestamp: ', msg['timestamp'])
# NOTE : Nothing recorded in world state. Uncomment to test it out.
# if world_state.number_of_observations_since_last_state > 0:
# timestamp = world_state.observations[-1].timestamp
# msg = world_state.observations[-1].text
# obs = json.loads(msg)
# print("{'timestamp': timestamp, 'observations': obs}")
# Video Frames
while world_state.number_of_video_frames_since_last_state < 1 and world_state.is_mission_running:
logger.info("Waiting for frames...")
time.sleep(0.05)
world_state = agent_host.getWorldState()
logger.info("Got frame!")
# import ipdb; ipdb.set_trace
# print('observations', world_state.observations)
# world_state.observations
if world_state.is_mission_running:
# timestamp = world_state.observations[-1].timestamp
# msg = world_state.observations[-1].text
# print(timestamp)
# print(msg)
frame = world_state.video_frames[-1]
img = Image.frombytes('RGB', (640, 480), bytes(frame.pixels))
# imageio.imsave("./tmp_imgs/{}.png".format(img_counter), img)
img_counter += 1
logger.info("Mission has stopped.")
time.sleep(1) # let the Mod recover
<AgentQuitFromReachingPosition>
<Marker tolerance="0.1" x="0.5" y="227" z="''' + str(
PATH_LENGTH + 0.5) + '''"/>
</AgentQuitFromReachingPosition>
</AgentHandlers>
</AgentSection>
</Mission>'''
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w',
0) # flush print output immediately
validate = True
agent_host = MalmoPython.AgentHost()
agent_host.addOptionalIntArgument(
"length,l", "Number of steps required to reach goal square.", 10)
# Eg set the length to 0 or 1 to test https://github.com/Microsoft/malmo/issues/23
agent_host.addOptionalFlag("stop,s", "Stop after required number of steps.")
# Eg if length is set to 10, will send 10 move commands and then wait for the mission to end.
# This can be used to test that commands are all being acted on, regardless of the speed they are sent at,
# and can give some indication of the latency between sending the final command and receiving the mission ended message.
agent_host.addOptionalFloatArgument(
"wait,w", "Number of seconds to wait between sending commands.", 0.1)
# Setting this to something slow (eg 0.1) should show a clear cycle of commands/observations/rewards,
# and a quit triggered after the correct number of commands.
# Setting this to something faster (eg 0.05) should still show a clear cylce of commands/observations/rewards,
(world_x - 1 - curr_x + 0.5 - curr_radius) * scale,
(world_y - 1 - curr_y + 0.5 - curr_radius) * scale,
(world_x - 1 - curr_x + 0.5 + curr_radius) * scale,
(world_y - 1 - curr_y + 0.5 + curr_radius) * scale,
outline="#fff",
fill="#fff")
self.root.update()
if sys.version_info[0] == 2:
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w',
0) # flush print output immediately
else:
import functools
print = functools.partial(print, flush=True)
agent_host = MalmoPython.AgentHost()
# add some args
agent_host.addOptionalStringArgument(
'mission_file', 'Path/to/file from which to load the mission.',
'./cliff_walking_1.xml')
agent_host.addOptionalFloatArgument('alpha',
'Learning rate of the Q-learning agent.',
0.1)
agent_host.addOptionalFloatArgument(
'epsilon', 'Exploration rate of the Q-learning agent.', 0.1)
agent_host.addOptionalFloatArgument('gamma', 'Discount factor.', 1.0)
agent_host.addOptionalFlag('load_model', 'Load initial model from model_file.')
agent_host.addOptionalStringArgument('model_file',
'Path to the initial model file', '')
agent_host.addOptionalFlag('debug', 'Turn on debugging.')
def __init__(self, mission_file):
super(MinecraftEnvRLKit, self).__init__()
self.agent_host = MalmoPython.AgentHost()
assets_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)),
'../assets')
mission_file = os.path.join(assets_dir, mission_file)
xml = Path(mission_file).read_text()
if not xml.startswith('<Mission'):
i = xml.index("<Mission")
xml = xml[i:]
self.ns = '{http://ProjectMalmo.microsoft.com}'
self.xml = etree.fromstring(xml)
self.load_mission_file(mission_file)
self.client_pool = None
self.mc_process = None
self.screen = None
self.image_obs = None
obs_from_grid = self.xml.findall(
'.//' + self.ns + 'ObservationFromGrid')[0].find(self.ns + 'Grid')
low = dict(obs_from_grid[0].attrib)
high = dict(obs_from_grid[1].attrib)
self.x_lim = (int(low['x']), int(high['x']))
self.y_lim = (int(low['y']), int(high['y']))
self.z_lim = (int(low['z']), int(high['z']))
self.x = int(high['x']) - int(low['x']) + 1
self.y = int(float(high['y']) - float(low['y']) + 1)
self.z = int(high['z']) - int(low['z']) + 1
self.full_dim = full_dim = self.z #self.x * self.y * self.z
self.partial_dim = 6
#self.partial_obs_space = Box(low=0, high=10, shape=(6,), dtype=np.float32)
self.obs_space = Box(low=0,
high=10,
shape=(full_dim + 4 + 6, ),
dtype=np.float32)
self.goal_space = Box(low=0,
high=10,
shape=(full_dim, ),
dtype=np.float32)
self.achieved_goal_space = Box(low=0,
high=10,
shape=(full_dim, ),
dtype=np.float32)
# self._state_goal = np.zeros((self.y, self.z, self.x))
# for y in range(1):
# #self._state_goal[y, 0, :] = BLOCK_KEY['brick_block']
# #self._state_goal[y, -1, :] = BLOCK_KEY['brick_block']
# #self._state_goal[y, :, 0] = BLOCK_KEY['brick_block']
# #self._state_goal[y, 3, -1] = BLOCK_KEY['brick_block']
# self._state_goal[y, :, -1] = BLOCK_KEY['brick_block']
# self._state_goal = self._state_goal.flatten()
self._state_goal = np.ones((full_dim, ))
self.last_obs = None
self.fix_goal = True
self.init_state = None
recordingsDirectory = "EatingRecordings"
try:
os.makedirs(recordingsDirectory)
except OSError as exception:
if exception.errno != errno.EEXIST: # ignore error if already existed
raise
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w',
0) # flush print output immediately
validate = True
# Create a pool of Minecraft Mod clients.
# By default, mods will choose consecutive mission control ports, starting at 10000,
# so running four mods locally should produce the following pool by default (assuming nothing else
# is using these ports):
my_client_pool = MalmoPython.ClientPool()
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10000))
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10001))
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10002))
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10003))
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print 'ERROR:', e
print agent_host.getUsage()
exit(1)
if agent_host.receivedArgument("help"):
print agent_host.getUsage()
exit(0)
<ModifierList type="deny-list"> <!-- Example deny-list: prevent agent from strafing -->
<command>strafe</command>
</ModifierList>
</ContinuousMovementCommands>
<ChatCommands />
<AgentQuitFromTouchingBlockType>
<Block type="redstone_block" />
</AgentQuitFromTouchingBlockType>
</AgentHandlers>
</AgentSection>
</Mission>'''
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w',
0) # flush print output immediately
agent_host = MalmoPython.AgentHost()
agent_host.addOptionalIntArgument(
"role,r", "For multi-agent missions, the role of this agent instance", 0)
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print 'ERROR:', e
print agent_host.getUsage()
exit(1)
if agent_host.receivedArgument("help"):
print agent_host.getUsage()
exit(0)
role = agent_host.getIntArgument("role")
print "Will run as role", role
tau = t - self.n + 1
# self.update_q_table(tau, s, next_a, current_r, T)
self.update_q_table(tau, s, next_a, current_r, T)
print(
"\n-------------------------------------- END ITERATION --------------------------------------\n"
)
self.observation_space = ['stone'] * num_doors
if __name__ == '__main__':
print('Starting...', flush=True)
expected_reward = 3390
my_client_pool = MalmoPython.ClientPool()
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10000))
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print('ERROR:', e)
print(agent_host.getUsage())
exit(1)
if agent_host.receivedArgument("help"):
print(agent_host.getUsage())
exit(0)
retry = 0
epocs = 1000
monty = Monty()
def __init__(self):
'''Initializes the class.'''
self.agent_host = MalmoPython.AgentHost()
self.root = Tk()
self.root.wm_title("Human Action Component")
human_agent_host.parse(sys.argv)
except RuntimeError as e:
print('ERROR:', e)
print(human_agent_host.getUsage())
exit(1)
if human_agent_host.receivedArgument("help"):
print(human_agent_host.getUsage())
exit(0)
my_role = human_agent_host.getIntArgument("role")
xml_filename = human_agent_host.getStringArgument("mission_xml")
if not xml_filename == "":
# load the mission from the specified XML file
my_mission = MalmoPython.MissionSpec(open(xml_filename).read(), True)
my_mission.requestVideo(640, 480)
my_mission.timeLimitInSeconds(300)
my_mission_record = MalmoPython.MissionRecordSpec(
'./hac_saved_mission.tgz')
my_mission_record.recordCommands()
my_mission_record.recordMP4(20, 400000)
my_mission_record.recordRewards()
my_mission_record.recordObservations()
human_agent_host.runMission(my_mission, my_mission_record, role=my_role)
else:
# create some sample missions
def load_mission_xml(self, mission_xml):
self.mission_spec = MalmoPython.MissionSpec(mission_xml, True)
logger.info("Loaded mission: " + self.mission_spec.getSummary())
<MissionQuitCommands/>
<RewardForCollectingItem>
<Item type="emerald" reward="1"/>
</RewardForCollectingItem>
</AgentHandlers>
</AgentSection>
</Mission>'''
if sys.version_info[0] == 2:
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w',
0) # flush print output immediately
else:
import functools
print = functools.partial(print, flush=True)
agent_host = MalmoPython.AgentHost()
agent_host.addOptionalStringArgument("savesDir,s",
"Location of Minecraft saves folder", "")
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print('ERROR:', e)
print(agent_host.getUsage())
exit(1)
if agent_host.receivedArgument("help"):
print(agent_host.getUsage())
exit(0)
saveFolders = []
savesDir = agent_host.getStringArgument("savesDir")
def init(self,
client_pool=None,
start_minecraft=None,
continuous_discrete=True,
add_noop_command=None,
max_retries=90,
retry_sleep=10,
step_sleep=0.001,
skip_steps=0,
videoResolution=None,
videoWithDepth=None,
observeRecentCommands=None,
observeHotBar=None,
observeFullInventory=None,
observeGrid=None,
observeDistance=None,
observeChat=None,
allowContinuousMovement=None,
allowDiscreteMovement=None,
allowAbsoluteMovement=None,
recordDestination=None,
recordObservations=None,
recordRewards=None,
recordCommands=None,
recordMP4=None,
gameMode=None,
forceWorldReset=None,
image_obs=False):
self.max_retries = max_retries
self.retry_sleep = retry_sleep
self.step_sleep = step_sleep
self.skip_steps = skip_steps
self.forceWorldReset = forceWorldReset
self.continuous_discrete = continuous_discrete
self.add_noop_command = add_noop_command
self.image_obs = image_obs
if videoResolution:
if videoWithDepth:
self.mission_spec.requestVideoWithDepth(*videoResolution)
else:
self.mission_spec.requestVideo(*videoResolution)
if observeRecentCommands:
self.mission_spec.observeRecentCommands()
if observeHotBar:
self.mission_spec.observeHotBar()
if observeFullInventory:
self.mission_spec.observeFullInventory()
if observeGrid:
self.mission_spec.observeGrid(*(observeGrid + ["grid"]))
if observeDistance:
self.mission_spec.observeDistance(*(observeDistance + ["dist"]))
if observeChat:
self.mission_spec.observeChat()
if allowContinuousMovement or allowDiscreteMovement or allowAbsoluteMovement:
# if there are any parameters, remove current command handlers first
self.mission_spec.removeAllCommandHandlers()
if allowContinuousMovement is True:
self.mission_spec.allowAllContinuousMovementCommands()
elif isinstance(allowContinuousMovement, list):
for cmd in allowContinuousMovement:
self.mission_spec.allowContinuousMovementCommand(cmd)
if allowDiscreteMovement is True:
self.mission_spec.allowAllDiscreteMovementCommands()
elif isinstance(allowDiscreteMovement, list):
for cmd in allowDiscreteMovement:
self.mission_spec.allowDiscreteMovementCommand(cmd)
if allowAbsoluteMovement is True:
self.mission_spec.allowAllAbsoluteMovementCommands()
elif isinstance(allowAbsoluteMovement, list):
for cmd in allowAbsoluteMovement:
self.mission_spec.allowAbsoluteMovementCommand(cmd)
if start_minecraft:
# start Minecraft process assigning port dynamically
self.mc_process, port = minecraft_py.start()
logger.info(
"Started Minecraft on port %d, overriding client_pool.", port)
client_pool = [('127.0.0.1', port)]
if client_pool:
if not isinstance(client_pool, list):
raise ValueError(
"client_pool must be list of tuples of (IP-address, port)")
self.client_pool = MalmoPython.ClientPool()
for client in client_pool:
self.client_pool.add(MalmoPython.ClientInfo(*client))
# TODO: produce observation space dynamically based on requested features
self.video_height = self.mission_spec.getVideoHeight(0)
self.video_width = self.mission_spec.getVideoWidth(0)
self.video_depth = self.mission_spec.getVideoChannels(0)
#self.observation_space = spaces.Box(low=0, high=255,
# shape=(self.video_height, self.video_width, self.video_depth))
self.observation_space = Dict([
('observation', self.obs_space),
('desired_goal', self.goal_space),
('achieved_goal', self.achieved_goal_space),
('state_observation', self.obs_space),
('state_desired_goal', self.goal_space),
('state_achieved_goal', self.achieved_goal_space),
])
# dummy image just for the first observation
self.last_image = np.zeros(
(self.video_height, self.video_width, self.video_depth),
dtype=np.uint8)
self._create_action_space()
# mission recording
self.mission_record_spec = MalmoPython.MissionRecordSpec(
) # record nothing
if recordDestination:
self.mission_record_spec.setDestination(recordDestination)
if recordRewards:
self.mission_record_spec.recordRewards()
if recordCommands:
self.mission_record_spec.recordCommands()
if recordMP4:
self.mission_record_spec.recordMP4(*recordMP4)
if gameMode:
if gameMode == "spectator":
self.mission_spec.setModeToSpectator()
elif gameMode == "creative":
self.mission_spec.setModeToCreative()
elif gameMode == "survival":
logger.warn(
"Cannot force survival mode, assuming it is the default.")
else:
assert False, "Unknown game mode: " + gameMode
<Height>''' + str(video_height) + '''</Height>
</ColourMapProducer>
<ContinuousMovementCommands turnSpeedDegs="180" />
</AgentHandlers>
</AgentSection>
</Mission>'''
if sys.version_info[0] == 2:
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w',
0) # flush print output immediately
else:
import functools
print = functools.partial(print, flush=True)
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print('ERROR:', e)
print(agent_host.getUsage())
exit(1)
if agent_host.receivedArgument("help"):
print(agent_host.getUsage())
exit(0)
agent_host.setVideoPolicy(MalmoPython.VideoPolicy.LATEST_FRAME_ONLY)
if agent_host.receivedArgument("test"):
num_reps = 1
else:
</VideoProducer>
<DiscreteMovementCommands />
<ChatCommands />
<RewardForSendingMatchingChatMessage>
<ChatMatch reward="100.0" regex="''' + items[
obj_id] + '''" description="Anything that matches the object."/>
</RewardForSendingMatchingChatMessage>
<RewardForSendingCommand reward="-1"/>
</AgentHandlers>
</AgentSection>
</Mission>
'''
# Create default Malmo objects:
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print('ERROR:', e)
print(agent_host.getUsage())
exit(1)
if agent_host.receivedArgument("help"):
print(agent_host.getUsage())
exit(0)
my_mission = MalmoPython.MissionSpec(mission_xml, True)
my_mission_record = MalmoPython.MissionRecordSpec("chat_reward.tgz")
# Attempt to start a mission:
max_retries = 3
</AgentHandlers>
</AgentSection>
</Mission>'''
recordingsDirectory="MovingTargetRecordings"
try:
os.makedirs(recordingsDirectory)
except OSError as exception:
if exception.errno != errno.EEXIST: # ignore error if already existed
raise
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0) # flush print output immediately
validate = True
my_client_pool = MalmoPython.ClientPool()
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10000))
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse( sys.argv )
except RuntimeError as e:
print 'ERROR:',e
print agent_host.getUsage()
exit(1)
if agent_host.receivedArgument("help"):
print agent_host.getUsage()
exit(0)
if agent_host.receivedArgument("test"):
num_reps = 1
import MalmoPython
import time
__author__ = "Chang Gao"
__copyright__ = "Copyright (c) 2017 Malmactor"
__license__ = "MIT"
import src.movement
from src.SuperMarioBros import *
layout = layout_fromdefault()
missionXML = layout_toxml(
layout, {"template_path": "SuperMarioBros/mission_template.xml"})
# start mission
agent_host = MalmoPython.AgentHost()
agent_host.addOptionalIntArgument("speed,s", "Length of tick, in ms.", 1)
my_mission = MalmoPython.MissionSpec(missionXML, True)
my_mission_record = MalmoPython.MissionRecordSpec()
my_mission.allowAllAbsoluteMovementCommands()
my_mission.allowAllDiscreteMovementCommands()
agent_host.startMission(my_mission, my_mission_record)
# Loop until mission starts:
world_state = agent_host.getWorldState()
while not world_state.has_mission_begun:
time.sleep(0.1)
world_state = agent_host.getWorldState()
for error in world_state.errors:
print "Error:", error.text
def main():
global NUM_REPS, N, ALPHA, GAMMA, MSE, agent
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w',
0) # flush print output immediately
debug = True if len(
sys.argv) > 1 and sys.argv[1] == "debug=True" else False
# Create default Malmo objects:
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print 'ERROR(Runtime):', e, "\n", agent_host.getUsage()
exit(1)
if agent_host.receivedArgument("help"):
print agent_host.getUsage()
exit(0)
# Set the size of the matrix
x, y = 21, 21
map_gen = environment_generator()
visual = visualization(x, y, debug)
for i in range(NUM_REPS):
print "Survival Mode # " + str(i + 1)
ws_interpre = world_state_interpreter(x, y)
action_available = action(agent_host)
# Attempt to start a mission:
max_retries = 3
for retry in range(max_retries):
try:
missionXML = map_gen.generator(False)
if i == 0:
missionXML = map_gen.generator(True)
my_mission = MalmoPython.MissionSpec(missionXML, True)
my_mission_record = MalmoPython.MissionRecordSpec()
agent_host.startMission(my_mission, my_mission_record)
time.sleep(3)
break
except RuntimeError as e:
if retry == max_retries - 1:
print "Error starting mission:", e
exit(1)
else:
time.sleep(2)
# Loop until mission starts:
print "Waiting for the mission to start ",
world_state = agent_host.getWorldState()
while not world_state.has_mission_begun:
sys.stdout.write(".")
time.sleep(0.1)
world_state = agent_host.getWorldState()
for error in world_state.errors:
print "Error:", error.text
print
print "Mission running "
show_best = False if (i + 1) % 5 != 0 else True
# Loop until mission ends:
while world_state.is_mission_running:
time.sleep(0.2)
matrix = None
ws_interpre.input(world_state)
ent_matrix = ws_interpre.entities_to_matrix()
env_matrix = ws_interpre.grid_matrix()
if ent_matrix and env_matrix:
visual.get_entities(ent_matrix)
visual.get_environment(env_matrix)
visual.draw()
matrix = visual.get_matrix()
if matrix:
agent.run(agent_host, matrix, False)
#time.sleep(0.1) # Use it for testing purpose
world_state = agent_host.getWorldState()
for error in world_state.errors:
print "Error:", error.text
MSE.append(agent.calculate_mse() * 100)
agent.reset_mse()
agent.replay(128)
print
print "Mission ended"
with open('Weights.txt', 'wb') as f:
pickle.dump(agent.get_weights(), f)
# Mission has ended.
time.sleep(1)
if debug:
visual.quit()
plt.plot(np.arange(len(MSE)), MSE, 'o')
m, b = np.polyfit(np.arange(len(MSE)), MSE, deg=1)
plt.plot(np.arange(len(MSE)), m * np.arange(len(MSE)) + b, '-')
plt.show()
canvas = tk.Canvas(root,
width=Constants.CANVAS_WIDTH,
height=Constants.CANVAS_HEIGHT,
borderwidth=0,
highlightthickness=0,
bg="white")
canvas.pack()
root.update()
validate = True
# Create a pool of Minecraft Mod clients.
# By default, mods will choose consecutive mission control ports, starting at 10000,
# so running four mods locally should produce the following pool by default (assuming nothing else
# is using these ports):
my_client_pool = MalmoPython.ClientPool()
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10000))
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10001))
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10002))
my_client_pool.add(MalmoPython.ClientInfo("127.0.0.1", 10003))
Constants.agent_host = MalmoPython.AgentHost()
try:
Constants.agent_host.parse(sys.argv)
except RuntimeError as e:
print 'ERROR:', e
print Constants.agent_host.getUsage()
exit(1)
if Constants.agent_host.receivedArgument("help"):
print Constants.agent_host.getUsage()
exit(0)
<Grid name="column">
<min x="0" y="-256" z="0" />
<max x="0" y="256" z="0" />
</Grid>
</ObservationFromGrid>
<VideoProducer want_depth="true">
<Width>''' + str(video_width) + '''</Width>
<Height>''' + str(video_height) + '''</Height>
</VideoProducer>
<AbsoluteMovementCommands />
</AgentHandlers>
</AgentSection>
</Mission>'''
validate = True
my_mission = MalmoPython.MissionSpec( missionXML, validate )
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse( sys.argv )
except RuntimeError as e:
print 'ERROR:',e
print agent_host.getUsage()
exit(1)
if agent_host.receivedArgument("help"):
print agent_host.getUsage()
exit(0)
agent_host.setObservationsPolicy(MalmoPython.ObservationsPolicy.LATEST_OBSERVATION_ONLY)
agent_host.setVideoPolicy(MalmoPython.VideoPolicy.LATEST_FRAME_ONLY)
def print_mission_spec():
my_mission = MalmoPython.MissionSpec()
print(my_mission.getAsXML(True))
# A sample that demonstrates a two-agent mission with discrete actions to dig and place blocks
import MalmoPython
import json
import logging
import math
import os
import random
import sys
import time
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w', 0) # flush print output immediately
# -- set up two agent hosts --
agent_host1 = MalmoPython.AgentHost()
agent_host2 = MalmoPython.AgentHost()
try:
agent_host1.parse( sys.argv )
except RuntimeError as e:
print 'ERROR:',e
print agent_host1.getUsage()
exit(1)
if agent_host1.receivedArgument("help"):
print agent_host1.getUsage()
exit(0)
# -- set up the mission --
xml = '''<?xml version="1.0" encoding="UTF-8" standalone="no" ?>
def visualize():
# Create default Malmo objects:
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print('ERROR:', e)
print(agent_host.getUsage())
exit(1)
if agent_host.receivedArgument("help"):
print(agent_host.getUsage())
exit(0)
if agent_host.receivedArgument("test"):
num_repeats = 1
else:
num_repeats = 10
for i in range(len(MISSION_FILES)):
# Attempt to start a mission:
my_mission = MalmoPython.MissionSpec(
open(MISSION_FILES[i + 3], 'r').read(), False)
"""planner = Agent(MISSION_START, MISSION_GOALS[i])
planner.create_world((MISSION_LOWER_BOUNDS[i][0], MISSION_UPPER_BOUNDS[i][0]),
(MISSION_LOWER_BOUNDS[i][1], MISSION_UPPER_BOUNDS[i][1]),
(MISSION_LOWER_BOUNDS[i][2], MISSION_UPPER_BOUNDS[i][2]),
MISSION_TEXT_FILES[i])
path = planner.astar()"""
if BOOL_DEBUG_PATHS:
w = dict()
for point in planner.world.walkable:
if point[1] in w.keys():
w[point[1]].add((point[0], point[2]))
else:
w[point[1]] = set()
w[point[1]].add([point[0], point[2]])
print(path)
# Comment this out when actually testing with the Malmo platform
#continue
max_retries = 3
for retry in range(max_retries):
try:
agent_host.startMission(my_mission,
MalmoPython.MissionRecordSpec())
break
except RuntimeError as e:
if retry == max_retries - 1:
print("Error starting mission:", e)
exit(1)
else:
time.sleep(2)
# Loop until mission starts:
print("Waiting for the mission to start ", )
world_state = agent_host.getWorldState()
while not world_state.has_mission_begun:
sys.stdout.write(".")
time.sleep(.5)
world_state = agent_host.getWorldState()
for error in world_state.errors:
print("Error:", error.text)
print()
print("Mission running ", )
# Loop until mission ends:
while world_state.is_mission_running:
sys.stdout.write(".")
time.sleep(1.0)
"""world_state = agent_host.getWorldState()
if path:
position = path.pop()
agent_host.sendCommand("tp {0} {1} {2}".format(*position))"""
print()
print("Mission ended")
outline=color_string,
fill=color_string)
if curr_x is not None and curr_y is not None:
self.canvas.create_oval(
(world_x - 1 - curr_x + 0.5 - curr_radius) * scale,
(world_y - 1 - curr_y + 0.5 - curr_radius) * scale,
(world_x - 1 - curr_x + 0.5 + curr_radius) * scale,
(world_y - 1 - curr_y + 0.5 + curr_radius) * scale,
outline="#fff",
fill="#fff")
self.root.update()
agent_host = MalmoPython.AgentHost()
# Find the default mission file by looking next to the schemas folder:
schema_dir = None
try:
schema_dir = os.environ['MALMO_XSD_PATH']
except KeyError:
print("MALMO_XSD_PATH not set? Check environment.")
exit(1)
mission_file = './random_map_3.xml'
print('')
print(os.path)
# add some args
agent_host.addOptionalStringArgument(
'mission_file', 'Path/to/file from which to load the mission.',
<Item reward="10" type="dirt"/>
</RewardForCollectingItem>
<RewardForDiscardingItem>
<Item reward="100" type="dirt"/>
</RewardForDiscardingItem>
</AgentHandlers>
</AgentSection>
</Mission>'''
my_mission = MalmoPython.MissionSpec(xml,True)
client_pool = MalmoPython.ClientPool()
client_pool.add( MalmoPython.ClientInfo('127.0.0.1',10000) )
client_pool.add( MalmoPython.ClientInfo('127.0.0.1',10001) )
MalmoPython.setLogging("", MalmoPython.LoggingSeverityLevel.LOG_OFF)
def safeStartMission(agent_host, mission, client_pool, recording, role, experimentId):
used_attempts = 0
max_attempts = 5
print("Calling startMission for role", role)
while True:
try:
agent_host.startMission(mission, client_pool, recording, role, experimentId)
break
except MalmoPython.MissionException as e:
errorCode = e.details.errorCode
if errorCode == MalmoPython.MissionErrorCode.MISSION_SERVER_WARMING_UP:
print("Server not quite ready yet - waiting...")
time.sleep(2)
elif errorCode == MalmoPython.MissionErrorCode.MISSION_INSUFFICIENT_CLIENTS_AVAILABLE:
import time
import json
import random
import math
import maze_gen2
from collections import defaultdict
DEFAULT_SIZE = 4
sys.stdout = os.fdopen(sys.stdout.fileno(), 'w',
0) # flush print output immediately
# Create default Malmo objects:
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print 'ERROR:', e
print agent_host.getUsage()
exit(1)
if agent_host.receivedArgument("help"):
print agent_host.getUsage()
exit(0)
#loading mission file
mission_file = "sarsaattempt1.xml"
with open(mission_file, 'r') as f:
print "Loading mission from %s" % mission_file
missionXML = f.read()