Added multi-agent support and created example agent for it

bin/Trainer.py

@@ -78,26 +78,24 @@ class Trainer(object):
     self._teams = [] # Team indexes for offensive and defensive teams
     self._SP = {} # Sever Parameters. Recieved when connecting to the server.
     self._isPlaying = False # Is a game being played?
-    self._agentPopen = None # Agent's process
+    self._agentPopen = [] # Agent's processes
     self.initMsgHandlers()
 
-  def launch_agent(self):
+  def launch_agent(self, player_num):
     """Launch the learning agent using the start.sh script and return a
     DummyPopen for the process.
     """
-    print '[Trainer] Launching Agent'
+    print '[Trainer] Launching Agent'+str(player_num)
     if self._agent_play_offense:
       assert self._numOffense > 0
       self._agentTeam = self._offenseTeam
-      self._agentNumInt = 1 if self._numOffense == 1 \
-                          else self._rng.randint(1, self._numOffense)
+      self._agentNumInt = player_num + 1
       numTeammates = self._numOffense - 1
       numOpponents = self._numDefense
     else:
       assert self._numDefense > 0
       self._agentTeam = self._defenseTeam
-      self._agentNumInt = 0 if self._numDefense == 1 \
-                          else self._rng.randint(0, self._numDefense)
+      self._agentNumInt = player_num
       numTeammates = self._numOffense
       numOpponents = self._numDefense - 1
     self._agentNumExt = self.convertToExtPlayer(self._agentTeam,
@@ -108,7 +106,7 @@ class Trainer(object):
                ' --playingOffense %i --serverPort %i'\
                %(self._agentTeam, self._agentNumExt, self._serverPort,
                  self._coachPort, self._logDir, numTeammates, numOpponents,
-                 self._agent_play_offense, self._agentServerPort)
+                 self._agent_play_offense, (self._agentServerPort - player_num))
     if self._record:
       agentCmd += ' --record'
     agentCmd = os.path.join(binary_dir, agentCmd)
@@ -640,8 +638,14 @@ class Trainer(object):
     """
     try:
       if self._agent:
-        self._agentPopen = self.launch_agent()
-        necProcesses.append([self._agentPopen,'agent'])
+        if self._agent_play_offense:
+          for i in xrange(self._numOffense):
+            self._agentPopen.append(self.launch_agent(i))
+            necProcesses.append([self._agentPopen[i], 'agent'+str(i)])
+        else:
+          for i in xrange(self._numDefense):
+            self._agentPopen.append(self.launch_agent(i))
+            necProcesses.append([self._agentPopen[i], 'agent'+str(i)])
       self.startGame()
       while self.checkLive(necProcesses):
         prevFrame = self._frame
@@ -653,8 +657,8 @@ class Trainer(object):
     except (KeyboardInterrupt, DoneError):
       print '[Trainer] Exiting'
     finally:
-      if self._agentPopen is not None:
-        self._agentPopen.send_signal(SIGINT)
+      for p in self._agentPopen:
+        p.send_signal(SIGINT)
       try:
         self._comm.sendMsg('(bye)')
       except:

example/example_high_level_random_agent.py

+#!/usr/bin/env python
+# encoding: utf-8
+
+# First Start the server: $> bin/start.py
+import random
+if __name__ == '__main__':
+  try:
+    from hfo import *
+  except:
+    print 'Failed to import hfo. To install hfo, in the HFO directory'\
+      ' run: \"pip install .\"'
+    exit()
+  # Connect 4 agents
+  agents = []
+  for i in xrange(4):
+    agents.append(hfo.HFOEnvironment())
+    agents[i].connectToAgentServer(6000-i, HFO_Features.HIGH_LEVEL_FEATURE_SET)
+  for episode in xrange(5):
+    status = HFO_Status.IN_GAME
+    while status == HFO_Status.IN_GAME:
+      # Grab the state features from the environment from the first agent
+      features = agents[0].getState()
+      # Take an action and get the current game status
+      for agent in agents:
+        rand_int = random.randint(0,3)
+        if rand_int == 0:
+          status = agent.act((HFO_Actions.MOVE, 0, 0))
+        elif rand_int == 1:
+          status = agent.act((HFO_Actions.SHOOT, 0, 0))
+        elif rand_int == 2:
+          status = agent.act((HFO_Actions.PASS, 0, 0))
+        elif rand_int == 3:
+          status = agent.act((HFO_Actions.DRIBBLE, 0, 0))
+    print 'Episode', episode, 'ended with',
+    # Check what the outcome of the episode was
+    if status == HFO_Status.GOAL:
+      print 'goal'
+    elif status == HFO_Status.CAPTURED_BY_DEFENSE:
+      print 'captured by defense'
+    elif status == HFO_Status.OUT_OF_BOUNDS:
+      print 'out of bounds'
+    elif status == HFO_Status.OUT_OF_TIME:
+      print 'out of time'
+    else:
+      print 'Unknown status', status
+      exit()
+  # Cleanup when finished
+  for agent in agents:
+    agent.cleanup()