Merge pull request #39 from drallensmith/docs_update

Feedback enabled

doc/manual.tex

@@ -324,6 +324,10 @@ features.
 \item [$3O$] {\textbf{X, Y, and Uniform Number of
     Opponents} - For each opponent: the x-position, y-position and
   uniform number of that opponent.}
+\item [$+1$] {\textbf{Last\_Action\_Success\_Possible} - Whether there is any chance
+	the last action taken was successful, either in accomplishing the
+	usual intent of the action or (primarily for the offense) in some other way such as
+	getting out of a goal-collision state. 1 for yes, -1 for no.}
 \end{enumerate}
 
 \begin{figure}[htp]
@@ -518,6 +522,10 @@ low-level features:
 	sorted by proximity to the agent.}
   \item [$O$]  {\textbf{Opponent Uniform Nums} [Unum] One uniform number for each opponent active in HFO,
 	sorted by proximity to the player.}
+  \item [$+1$]  {\textbf{Last\_Action\_Success\_Possible} [Boolean] Whether there is any chance
+		the last action taken was successful, either in accomplishing the
+		usual intent of the action or (primarily for the offense) in some other way such as getting
+		out of a goal-collision state.}
 \end{enumerate}
 
 \section{Action Space}

example/hand_coded_defense_agent.py

@@ -19,7 +19,7 @@ except ImportError:
     ' run: \"pip install .\"')
   exit()
 
-GOAL_POS_X = 1.0
+GOAL_POS_X = 0.9
 GOAL_POS_Y = 0.0
 
 # below - from hand_coded_defense_agent.cpp except LOW_KICK_DIST

hfo/hfo.py

@@ -6,11 +6,12 @@ import os
 hfo_lib = cdll.LoadLibrary(os.path.join(os.path.dirname(__file__),
                                         'libhfo_c.so'))
 
-''' Possible feature sets '''
+"""Possible feature sets"""
 NUM_FEATURE_SETS = 2
 LOW_LEVEL_FEATURE_SET, HIGH_LEVEL_FEATURE_SET = list(range(NUM_FEATURE_SETS))
 
-''' An enum of the possible HFO actions
+"""
+An enum of the possible HFO actions, including:
   [Low-Level] Dash(power, relative_direction)
   [Low-Level] Turn(direction)
   [Low-Level] Tackle(direction)
@@ -25,25 +26,50 @@ LOW_LEVEL_FEATURE_SET, HIGH_LEVEL_FEATURE_SET = list(range(NUM_FEATURE_SETS))
   [High-Level] Dribble(): Offensive dribble
   [High-Level] Catch(): Catch the ball (Goalie Only)
   NOOP(): Do Nothing
-  QUIT(): Quit the game '''
+  QUIT(): Quit the game
+"""
 NUM_HFO_ACTIONS = 20
-DASH, TURN, TACKLE, KICK, KICK_TO, MOVE_TO, DRIBBLE_TO, INTERCEPT, \
-    MOVE, SHOOT, PASS, DRIBBLE, CATCH, NOOP, QUIT, REDUCE_ANGLE_TO_GOAL,MARK_PLAYER,DEFEND_GOAL,GO_TO_BALL,REORIENT = list(range(NUM_HFO_ACTIONS))
-ACTION_STRINGS = ["Dash", "Turn", "Tackle", "Kick", "KickTo", "MoveTo", "DribbleTo", "Intercept", "Move", "Shoot", "Pass", "Dribble", "Catch", "No-op", "Quit", "Reduce_Angle_To_Goal", "Mark_Player", "Defend_Goal", "Go_To_Ball", "Reorient"]
-
-''' Possible game status
+DASH,TURN,TACKLE,KICK,KICK_TO,MOVE_TO,DRIBBLE_TO,INTERCEPT,MOVE,SHOOT,PASS,DRIBBLE,CATCH,NOOP,QUIT,REDUCE_ANGLE_TO_GOAL,MARK_PLAYER,DEFEND_GOAL,GO_TO_BALL,REORIENT = list(range(NUM_HFO_ACTIONS))
+ACTION_STRINGS = {DASH: "Dash",
+                  TURN: "Turn",
+                  TACKLE: "Tackle",
+                  KICK: "Kick",
+                  KICK_TO: "KickTo",
+                  MOVE_TO: "MoveTo",
+                  DRIBBLE_TO: "DribbleTo",
+                  INTERCEPT: "Intercept",
+                  MOVE: "Move",
+                  SHOOT: "Shoot",
+                  PASS: "Pass",
+                  DRIBBLE: "Dribble",
+                  CATCH: "Catch",
+                  NOOP: "No-op",
+                  QUIT: "Quit",
+                  REDUCE_ANGLE_TO_GOAL: "Reduce_Angle_To_Goal",
+                  MARK_PLAYER: "Mark_Player",
+                  DEFEND_GOAL: "Defend_Goal",
+                  GO_TO_BALL: "Go_To_Ball",
+                  REORIENT: "Reorient"}
+
+"""
+Possible game statuses:
   [IN_GAME] Game is currently active
   [GOAL] A goal has been scored by the offense
   [CAPTURED_BY_DEFENSE] The defense has captured the ball
   [OUT_OF_BOUNDS] Ball has gone out of bounds
   [OUT_OF_TIME] Trial has ended due to time limit
   [SERVER_DOWN] Server is not alive
-'''
+"""
 NUM_GAME_STATUS_STATES = 6
 IN_GAME, GOAL, CAPTURED_BY_DEFENSE, OUT_OF_BOUNDS, OUT_OF_TIME, SERVER_DOWN = list(range(NUM_GAME_STATUS_STATES))
-STATUS_STRINGS = ["InGame", "Goal", "CapturedByDefense", "OutOfBounds", "OutOfTime", "ServerDown"]
-
-''' Possible sides '''
+STATUS_STRINGS = {IN_GAME: "InGame",
+                  GOAL: "Goal",
+                  CAPTURED_BY_DEFENSE: "CapturedByDefense",
+                  OUT_OF_BOUNDS: "OutOfBounds",
+                  OUT_OF_TIME: "OutOfTime",
+                  SERVER_DOWN: "ServerDown"}
+
+"""Possible sides."""
 RIGHT, NEUTRAL, LEFT = list(range(-1,2))
 
 class Player(Structure): pass
@@ -109,7 +135,14 @@ class HFOEnvironment(object):
       play_goalie: is this player the goalie
       record_dir: record agent's states/actions/rewards to this directory
     """
-    hfo_lib.connectToServer(self.obj, feature_set, config_dir.encode('utf-8'), server_port,server_addr.encode('utf-8'), team_name.encode('utf-8'), play_goalie, record_dir.encode('utf-8'))
+    hfo_lib.connectToServer(self.obj,
+                            feature_set,
+                            config_dir.encode('utf-8'),
+                            server_port,server_addr.encode('utf-8'),
+                            team_name.encode('utf-8'),
+                            play_goalie,
+                            record_dir.encode('utf-8'))
+
   def getStateSize(self):
     """ Returns the number of state features """
     return hfo_lib.getStateSize(self.obj)

src/agent.h

+// -*-c++-*-
+
 #ifndef AGENT_H
 #define AGENT_H
 
@@ -63,6 +65,7 @@ protected:
   std::vector<float> params;           // Parameters of current action
   int num_teammates;                   // Number of teammates
   int num_opponents;                   // Number of opponents
+  bool last_action_status;  // Recorded return status of last action
 
  public:
   inline const std::vector<float>& getState() { return state; }
@@ -72,6 +75,7 @@ protected:
   int getUnum(); // Returns the uniform number of the player
   inline int getNumTeammates() { return num_teammates; }
   inline int getNumOpponents() { return num_opponents; }
+  inline bool getLastActionStatus() { return last_action_status; }
 
   inline void setFeatureSet(hfo::feature_set_t fset) { feature_set = fset; }
   inline std::vector<float>* mutable_params() { return &params; }

src/bhv_basic_move.cpp

@@ -65,7 +65,7 @@ Bhv_BasicMove::execute( PlayerAgent * agent )
     // tackle
     if ( Bhv_BasicTackle( 0.8, 80.0 ).execute( agent ) )
     {
-        return true;
+      return true;
     }
 
     const WorldModel & wm = agent->world();
@@ -84,16 +84,17 @@ Bhv_BasicMove::execute( PlayerAgent * agent )
     {
         dlog.addText( Logger::TEAM,
                       __FILE__": intercept" );
-        Body_Intercept().execute( agent );
+        bool success = Body_Intercept().execute( agent );
         agent->setNeckAction( new Neck_OffensiveInterceptNeck() );
-
-        return true;
+        return success;
     }
 
     const Vector2D target_point = Strategy::i().getPosition( wm.self().unum() );
     const double dash_power = Strategy::get_normal_dash_power( wm );
 
-    double dist_thr = wm.ball().distFromSelf() * 0.1;
+    const BallObject& ball = wm.ball();
+
+    double dist_thr = ball.distFromSelf() * 0.1;
     if ( dist_thr < 1.0 ) dist_thr = 1.0;
 
     dlog.addText( Logger::TEAM,
@@ -105,21 +106,27 @@ Bhv_BasicMove::execute( PlayerAgent * agent )
     agent->debugClient().setTarget( target_point );
     agent->debugClient().addCircle( target_point, dist_thr );
 
-    if ( ! Body_GoToPoint( target_point, dist_thr, dash_power
-                           ).execute( agent ) )
-    {
-        Body_TurnToBall().execute( agent );
+    bool success = false;
+
+    if ( Body_GoToPoint( target_point, dist_thr, dash_power
+			 ).execute( agent ) ||
+	 Body_TurnToBall().execute( agent ) ) {
+      if (ball.posValid() || wm.self().collidesWithPost()) {
+	success = true;
+      }
+    } else {
+      success = false;
     }
 
     if ( wm.existKickableOpponent()
-         && wm.ball().distFromSelf() < 18.0 )
-    {
-        agent->setNeckAction( new Neck_TurnToBall() );
-    }
-    else
-    {
-        agent->setNeckAction( new Neck_TurnToBallOrScan() );
+         && ball.distFromSelf() < 18.0 ) {
+      agent->setNeckAction( new Neck_TurnToBall() );
+    } else if ( ball.posValid() ) {
+      agent->setNeckAction( new Neck_TurnToBallOrScan() );
+    } else {
+      agent->setNeckAction( new Neck_TurnToBall() );
     }
 
-    return true;
+
+    return success;
 }

src/bhv_basic_move.h

@@ -36,7 +36,7 @@ public:
     Bhv_BasicMove()
       { }
 
-    bool execute( rcsc::PlayerAgent * agent );
+  bool execute( rcsc::PlayerAgent * agent );
 
 private:
     double getDashPower( const rcsc::PlayerAgent * agent );

src/common.hpp

@@ -33,15 +33,15 @@ enum action_t
   DRIBBLE,    // [High-Level] Dribble(): Offensive dribble
   CATCH,      // [High-Level] Catch(): Catch the ball (Goalie only!)
   NOOP,       // Do nothing
-  QUIT,        // Special action to quit the game
+  QUIT,       // Special action to quit the game
   REDUCE_ANGLE_TO_GOAL, // [High-Level] Reduce_Angle_To_Goal : Reduces the shooting angle
-  MARK_PLAYER, 			// [High-Level] Mark_Player(opponent_unum [0,11]) : Moves to the position in between the kicker and a given player
+  MARK_PLAYER, // [High-Level] Mark_Player(opponent_unum [0,11]) : Moves to the position in between the kicker and a given player
   DEFEND_GOAL,
   GO_TO_BALL,
   REORIENT  // [High-Level] Handle lost position of self/ball, misc other situations; variant of doPreprocess called in DRIBBLE
 };
 
-// Status of a HFO game
+// Status of an HFO game
 enum status_t
 {
   IN_GAME,             // Game is currently active
@@ -176,7 +176,7 @@ inline std::string ActionToString(action_t action) {
 };
 
 /**
- * Returns a string representation of a game_status.
+ * Returns a string representation of a game status.
  */
 inline std::string StatusToString(status_t status) {
   switch (status) {

src/feature_extractor.h

+// -*-c++-*-
+
 #ifndef FEATURE_EXTRACTOR_H
 #define FEATURE_EXTRACTOR_H
 
@@ -12,7 +14,8 @@ public:
   virtual ~FeatureExtractor();
 
   // Updated the state features stored in feature_vec
-  virtual const std::vector<float>& ExtractFeatures(const rcsc::WorldModel& wm) = 0;
+  virtual const std::vector<float>& ExtractFeatures(const rcsc::WorldModel& wm,
+						    bool last_action_status) = 0;
 
   // Record the current state
   void LogFeatures();

src/highlevel_feature_extractor.cpp

@@ -10,19 +10,21 @@ using namespace rcsc;
 HighLevelFeatureExtractor::HighLevelFeatureExtractor(int num_teammates,
                                                      int num_opponents,
                                                      bool playing_offense) :
-    FeatureExtractor(num_teammates, num_opponents, playing_offense)
+  FeatureExtractor(num_teammates, num_opponents, playing_offense)
 {
   assert(numTeammates >= 0);
   assert(numOpponents >= 0);
   numFeatures = num_basic_features + features_per_teammate * numTeammates
       + features_per_opponent * numOpponents;
+  numFeatures++; // action status
   feature_vec.resize(numFeatures);
 }
 
 HighLevelFeatureExtractor::~HighLevelFeatureExtractor() {}
 
-const std::vector<float>& HighLevelFeatureExtractor::ExtractFeatures(
-    const WorldModel& wm) {
+const std::vector<float>&
+HighLevelFeatureExtractor::ExtractFeatures(const rcsc::WorldModel& wm,
+					   bool last_action_status) {
   featIndx = 0;
   const ServerParam& SP = ServerParam::i();
   const SelfObject& self = wm.self();
@@ -178,6 +180,12 @@ const std::vector<float>& HighLevelFeatureExtractor::ExtractFeatures(
     addFeature(FEAT_INVALID);
   }
 
+  if (last_action_status) {
+    addFeature(FEAT_MAX);
+  } else {
+    addFeature(FEAT_MIN);
+  }
+
   assert(featIndx == numFeatures);
   // checkFeatures();
   return feature_vec;

src/highlevel_feature_extractor.h

+// -*-c++-*-
+
 #ifndef HIGHLEVEL_FEATURE_EXTRACTOR_H
 #define HIGHLEVEL_FEATURE_EXTRACTOR_H
 
@@ -18,7 +20,8 @@ public:
   virtual ~HighLevelFeatureExtractor();
 
   // Updated the state features stored in feature_vec
-  virtual const std::vector<float>& ExtractFeatures(const rcsc::WorldModel& wm);
+  virtual const std::vector<float>& ExtractFeatures(const rcsc::WorldModel& wm,
+						    bool last_action_status);
 
 protected:
   // Number of features for non-player objects.

src/lowlevel_feature_extractor.cpp

@@ -17,13 +17,15 @@ LowLevelFeatureExtractor::LowLevelFeatureExtractor(int num_teammates,
   numFeatures = num_basic_features +
       features_per_player * (numTeammates + numOpponents);
   numFeatures += numTeammates + numOpponents; // Uniform numbers
+  numFeatures++; // action state
   feature_vec.resize(numFeatures);
 }
 
 LowLevelFeatureExtractor::~LowLevelFeatureExtractor() {}
 
-const std::vector<float>& LowLevelFeatureExtractor::ExtractFeatures(
-    const WorldModel& wm) {
+const std::vector<float>&
+LowLevelFeatureExtractor::ExtractFeatures(const rcsc::WorldModel& wm,
+					  bool last_action_status) {
   featIndx = 0;
   const ServerParam& SP = ServerParam::i();
   // ======================== SELF FEATURES ======================== //
@@ -197,7 +199,7 @@ const std::vector<float>& LowLevelFeatureExtractor::ExtractFeatures(
       detected_teammates++;
     }
   }
-  // Add -2 features for any missing teammates
+  // Add -1 features for any missing teammates
   for (int i=detected_teammates; i<numTeammates; ++i) {
     addFeature(FEAT_MIN);
   }
@@ -212,11 +214,17 @@ const std::vector<float>& LowLevelFeatureExtractor::ExtractFeatures(
       detected_opponents++;
     }
   }
-  // Add -2 features for any missing opponents
+  // Add -1 features for any missing opponents
   for (int i=detected_opponents; i<numOpponents; ++i) {
     addFeature(FEAT_MIN);
   }
 
+  if (last_action_status) {
+    addFeature(FEAT_MAX);
+  } else {
+    addFeature(FEAT_MIN);
+  }
+
   assert(featIndx == numFeatures);
   checkFeatures();
   return feature_vec;

src/lowlevel_feature_extractor.h

+// -*-c++-*-
+
 #ifndef LOWLEVEL_FEATURE_EXTRACTOR_H
 #define LOWLEVEL_FEATURE_EXTRACTOR_H
 
@@ -12,7 +14,8 @@ public:
   virtual ~LowLevelFeatureExtractor();
 
   // Updated the state features stored in feature_vec
-  virtual const std::vector<float>& ExtractFeatures(const rcsc::WorldModel& wm);
+  virtual const std::vector<float>& ExtractFeatures(const rcsc::WorldModel& wm,
+						    bool last_action_status);
 
 protected:
   // Number of features for non-player objects.

src/sample_player.cpp

@@ -257,7 +257,7 @@ SamplePlayer::actionImpl()
       lastTrainerMessageTime = audioSensor().trainerMessageTime().cycle();
     }
     if (feature_extractor != NULL) {
-      feature_extractor->ExtractFeatures(this->world());
+      feature_extractor->ExtractFeatures(this->world(), true);
       feature_extractor->LogFeatures();
     }
   }