Added start script for agent. Refactored state variables.

bin/Trainer.py

@@ -85,14 +85,19 @@ class Trainer(object):
       self._agentTeam = self._offenseTeam
       self._agentNumInt = 1 if self._numOffense == 1 \
                           else self._rng.randint(1, self._numOffense)
+      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)
+      numTeammates = self._numOffense
+      numOpponents = self._numDefense - 1
     self._agentNumExt = self.convertToExtPlayer(self._agentTeam,
                                                 self._agentNumInt)
-    agentCmd = 'start_agent.sh -t %s -u %i'%(self._agentTeam, self._agentNumExt)
+    agentCmd = 'start_agent.sh -t %s -u %i --numTeammates %i --numOpponents %i'\
+               %(self._agentTeam, self._agentNumExt, numTeammates, numOpponents)
     agentCmd = agentCmd.split(' ')
     # Ignore stderr because librcsc continually prints to it
     kwargs = {'stderr':open('/dev/null','w')}
@@ -475,22 +480,23 @@ class Trainer(object):
 
   def getOffensiveResetPosition(self):
     """ Returns a random position for an offensive player. """
-    offsets = [
-      [-1,-1],
-      [-1,1],
-      [1,1],
-      [1,-1],
-      [0,2],
-      [0,-2],
-      [-2,-2],
-      [-2,2],
-      [2,2],
-      [2,-2],
-    ]
-    offset = offsets[self._rng.randint(len(offsets))]
-    offset_from_ball = 0.1 * self.PITCH_LENGTH * self._rng.rand(2) + \
-                       0.1 * self.PITCH_LENGTH * numpy.array(offset)
-    return self.boundPoint(self._ballPosition + offset_from_ball)
+    # offsets = [
+    #   [-1,-1],
+    #   [-1,1],
+    #   [1,1],
+    #   [1,-1],
+    #   [0,2],
+    #   [0,-2],
+    #   [-2,-2],
+    #   [-2,2],
+    #   [2,2],
+    #   [2,-2],
+    # ]
+    # offset = offsets[self._rng.randint(len(offsets))]
+    # offset_from_ball = 0.1 * self.PITCH_LENGTH * self._rng.rand(2) + \
+    #                    0.1 * self.PITCH_LENGTH * numpy.array(offset)
+    # return self.boundPoint(self._ballPosition + offset_from_ball)
+    return self._ballPosition
 
   def getDefensiveResetPosition(self):
     """ Returns a random position for a defensive player. """

src/agent.h

@@ -37,8 +37,7 @@
 class Agent : public rcsc::PlayerAgent {
 public:
   Agent();
-  virtual ~Agent();
-  std::vector<float> getState();
+  virtual ~Agent() {};
   virtual FieldEvaluator::ConstPtr getFieldEvaluator() const;
 
 protected:
@@ -59,11 +58,24 @@ protected:
   virtual FieldEvaluator::ConstPtr createFieldEvaluator() const;
   virtual ActionGenerator::ConstPtr createActionGenerator() const;
 
- private:
+  // Updated the state features stored in feature_vec
+  void updateStateFeatures();
+
   // Add the angle and distance to the landmark to the feature_vec
   void addLandmarkFeature(const rcsc::Vector2D& landmark,
-                          const rcsc::Vector2D& self_pos,
-                          std::vector<float>& feature_vec);
+                          const rcsc::Vector2D& self_pos);
+
+  int numTeammates;
+  int numOpponents;
+  int numFeatures; // Total number of features
+  // Number of features for non-player objects. Clearly this is the answer.
+  const static int num_basic_features = 42;
+  // Number of features for each player or opponent in game.
+  const static int features_per_player = 5;
+  std::vector<float> feature_vec; // Contains the current features
+  int featIndx; // Feature being populated
+
+ private:
   bool doPreprocess();
   bool doShoot();
   bool doForceKick();