David H. JohnsonSimulated Robot Soccer Assignment4/16/98THE TASKThe purpose of this assignment is to create an intelligent soccer team.  Theteams must be executable in either the JavaSoccer or ASCIISoccer simulationenvironments.  Information for these packages can be found at their respectivesites:http://www.cc.gatech.edu/grads/b/Tucker.Balch/JavaBots/EDU/gatech/cc/is/docs/index.htmlandhttp://www.cc.gatech.edu/grads/b/Tucker.Balch/soccer/The soccer players must posses some intelligence in order to play a good gameof soccer, and this intelligence can be implemented in several ways.  Theplayers can be either homogeneous (ie the SchemaDemo team) or heterogenous(ie the BasicTeam).  On a homogeneous team, each member is attempting to dothe same thing, only from a different perspective (ie different sensor readings).  Whereas with a heterogenous team each member is programmed to doa specific task and work together as a team.  I chose a heterogenous approachto the team.The individual team members can be classified by what type of intelligent agentthey are:  Reflexive, State-Based, or Goal-Oriented.  These are the threebasic intelligent agent types introduced in Russell and Norvig's _ArtificialIntelligence:  A Modern Approach_.  The reflexive agent is one that takes sensory input and reacts directly on that input.  Each time it receives newinput it responds without any reference goals or previous experience.  Slightlymore complex than reflex agents are state-based agents.  These have previousstate knowledge such as "i was just trying to intercept a pass" and use thisknowledge along with the current sensor input and a world model to choose anaction to take.  The final agent type is goal-oriented.  Goal-oriented agentshave a specific goal to accomplish, and attempt to predict the effects of possible actions to determine which action is most useful for reachingthe goal.  My team members are reflexive agents.A final classification of intelligent agents is whether or not they learn.The intelligence can be hard-coded (through heuristics) or they can learn fromprevious experiences.  The robots for this assignment do not learn.  They havehard-coded heuristics.Reflexive, heuristic, and heterogenous robots were chosen to hopefully showthat relatively dumb robots working together can perform well.IMPLEMENTATIONGlobal variables as listed in the code are values determined at the beginningof each simulation step.  Except for Dec2 move which is used to determinewhere to move after each step, and boolean kickit which is true if the robotwants to try to kick the ball.  The constants, SIDE and ME are set in theConfiguration() method of each robot.The team members are unique in the role they play on the team and aredetermined by their robot ID.  They are:Goalie, Offside, Designated Driver, Backup, and Center.   - Goalie -The goalie is responsible for keeping itself between the ball and the goalat all times.  The goalie must also deflect a ball if it is dangerously closeto the goal.  If the goalie is too far from the goal, or outside the widthof the goal, it will return to the center as quickly as possible.  Otherwiseit will try to keep the ball from going north or south along the y axis.  Itstops when the ball is sufficiently close to its x-axis to avoid moving past theball.  The goalie moves toward the wall of the goal at all times.  If the ballis behind the goalie the goalie will chase it down and kick it out as quicklyas possible.  All the code for this is contained in a single function, play_goalie(), which needs no helpers.   - Offside -This player targets the opponent who is closest to the opponents goal.  Oncethe opponent is targeted "Offside" calculates and goes directly toward thepoint directly behind the opponent toward his goal.  On the way to this point"Offside" avoids collision with any players other than the target.  Offsideserves two purposes.  One is to serve as a blocker to keep the target fromrushing an incoming teammate driving to score.  Another is to effectivelycombat the traditional goalie.  Two important functions are introduced inthis player's logic.1) Vec2 closest_to( Vec2 point, Vec2[] objects)	Returns which of the objects is closest to a given point.  The distance	values are calculated using basic vector arithmetic.  In this case it	is used to determine which opponent is closest to the opponents goal.	It is also used to calculate which teammate is closest to the ball, 	which teammate is closest to the robot, and which opponent is closest 	to the robot.2) void avoidcollision()	This is more like a get out of the way if you've collided.  If you	are close enough to a teammate or enemy, then turn in the opposite	direction.  This is not used when in control of the ball (driving to	the goal) or by the goalie.   - Designated Driver -The Designated Driver tries to continually drive the ball.  Using the functiondrive_ball().void drive_ball()	If the robot is behind the ball with respect to the goal (ie a line	can be drawn from the robot to the goal nearly passing through the	ball) and the ball is fairly close, then move toward the goal.  If	the robot can move toward the goal and is relatively close to the goal	then try to kick the ball.  If you aren't behind the ball, then	get behind the ball, and avoid collision while trying.drive_ball() uses two functions behind_point() and get_behind():boolean behind_point( Vec2 point, Vec2 orient)	Determines if you are behind a point and oriented toward another point.	This is determined by calculating how much difference there is in	the angle of the point and the angle of the orientation.  If you are	directly behind the point there should be no difference, but a little	leeway is given.void get_behind( Vec2 point, Vec2 orient)	Attempts to move behind a point with respect to an orientation point.	The robot wants the point between itself and the orientation.  To	see how this is calculated picture a line from the orientation through	the point, and line from the robot through the point.  Where these	lines emerge an angle, theta is formed.  If the cosine of the angle	is positive you are in front of the point, if it is negative you are	behind the point.  If you are behind the point, then the sine of theta	will be positive if you are to the left of the point, and negative	if you are to the right.  Also calculate a vector from the point away	from the orientation to the point you would ideally like to go.  If	you are behind the point, go strait there.  If you are behind the	point and on the left or right, then rotate the behind vector 90	degrees left or right to head around the point.  Calculating theta	and the behind vector are trivial matters.   - Backup - This robot calculates a spot three robot radii from the ball towards his ownside and tries to get behind it with respect to the opponents goal.  If "Backup" is the closest robot to the ball it will try to drive the ball.  Thisserves as a backup for the designated driver, hence the name "Backup".   - Center -This robot tries to get behind the center of the field and the opponent's goal.If it is the closest robot to the ball then it tries to drive the ball.  Themain purpose of this robot is to sit at center field and wait for the fastbreak to come.  At first I was a little queasy about implementing this robot.But decided that 1) Prof. Atkeson specifically said there were no rules, and2) this is a reflexive robot.  Because reflexive robots are dumb they haveto do everything they can to take advantage of their environment.  This isthe way animals survive with natural selection.  They take advantage of theenvironment, and if they don't, they die.  So the center is doing whatit can to take advantage of the environment for the benefit of the team.PERFORMANCE/IMPROVEMENTSOf course the DTeam consistently beats all of the stock teams, including Brian McNamara's well designed team.  This can be expected from most new teams considering the major testing most teams will do is going to be against thestock teams.  I created another team, called Team-Brick.  The name describestheir functionality, as they just stay in their start position.  Every oneof the stock teams stalled the game after at most one shot was taken.  Wherethe lineup put Team-Brick near the ball.  The other teams couldn't keep fromrunning into each other getting to the ball, or were programmed to stay toofar away from opponents to get to the ball.  The small, but effective collisionavoidance method should help DTeam perform well.  I believe the reason collision detection effective is that the simulation is so dynamic.  Mostevery obstacle is moving so a slight push away from a collision gives freedomfrom the obstacle.  And after you have freedom from the obstacle your newheading is probably changed as well, meaning you probably won't bump into theobstacle again, unless you are heading for the same place (ie behind theball).  Nevertheless the collision avoidance was quite sufficient to get theoffsides man back to the goalie quickly.However, by no means is DTeam perfect.  There are several things I would liketo change and improve with the DTeam.  First the get_behind() function couldbe improved.  The way it works now, it is practically a planning mechanism.  Although a primitive one.  The robots have to do too much calculating toget behind an object.  Using the goalie, and "offside" along with threeSchemaDemo (or BrianTeam) members would be more effective.  The idea of ahomogeneous goal striking team is appealing.  I would have liked to look atClay more.  It would be nice to have two specialized players and a 3 robotstriking team that could be trained with Schema weights.  The get_behind()and behind_point functions are good because they can be reused to improvethe team, along with a function that I wrote, but didn't integrate callednearpath().  Nearpath() finds the closest to the path from you to a goal.  Ifnearpath() found that there was something blocking your shot you could setup to get behind the ball toward a teammate for a pass.  However, this may diminish the effectiveness of driving the ball.  Another thing you could do is just check to see if you were ever behind_point()the ball relative to a teammate for a pass.  Kicking is always better than dribbling because the ball moves faster and can shake the opponents.  The onlyproblem would be that unless the team was completely homogeneous you would needsome sort of communications between robots to make sure you don't passit to the goalie or the "offside".  The communication might take too long forit to be effective.  The robots in soccer have to react immediately to carryout a strategy.CREDITSI used the BasicTeam.java skeleton to create a heterogenous team.