Here we answer various questions regarding the 3D simulator and the UT Austin Villa code base.

Is there something missing? Something you want to add? Email us, and we'll add it.

• Are there any example agents I can test my agent against? Yes, you can play against some binaries from the 2012 competiton. You will find them in the /projects/cs344M.pstone/3d/opponents/ directory. These binaries include the top 4 teams in the tournament and all have been configured to run on UTCS. The full set of RoboCup 2012 released binaries are located at http://simspark.sourceforge.net/binaries/RoboCup2012/, however many will of them will not run on UTCS (they all should work out of the box on a vanilla install of Fedora 16 64-bit and I got most of them to work on Ubuntu 11.10). To see how the different teams did during the tournament you can check the RoboCup 2012 3D simulation league results page.
  General instructions for running a binary:
  • Most binaries will only work on a 64-bit machine so make sure you are running on a 64-bit machine. To check this run the following command and make sure it returns a value of 64.

         % getconf LONG_BIT
    

    If you're not a 64-bit machine you can get a list of 64-bit machines to ssh into and run a team on with the following command

         % cshosts pub64
    

  • You can run any of the teams from within their agent directory with the following command:

         % ./start.sh localhost
    

    If you are running a server on a remote machine (usually a good idea when running a full game) Replace localhost with the name of the machine the server is running on.

         % ./start.sh remote_machine_name
    

  • Important: Some of the teams don't kill themselves when the server is shutdown meaning that you need to manually kill the team after running it. This is true of the korachikoalas team and possibly others. For these teams you need to manually kill their agent processes when you're finished running them.

         % ./kill.sh
    

• Is there a good place to find out what the UT Austin Villa team does for different things? There is a farily comprehensive description of the UT Austin Villa team in the UT Austin Villa 2011 3D Simulation Team Report.
• How can I run the server remotely on UTCS but watch games locally on my machine at home?
  • 1. Follow the RoboViz installation instructions (https://sites.google.com/site/umroboviz/usage/setup - ie. download JDK, checkout from repo, compile with sh build-<OS>.sh)
  • 2. Modify the 'Server Host' line in your Resources/config.txt file to point to the UT Unix machine that you're running the server on
  • 3. Run server on remote machine as usual (run_server3d)
  • 4. Run roboviz monitor on local machine (./roboviz.sh)
  • 5. Run agents remotely on another UTCS machine as usual
• How do I use the UT Austin Villa positioning system? In order to use dynamic role assignment with the positioning system specified in the paper Positioning to Win: A Dynamic Role Assignment and Formation Positioning System you will need to set the useDynamicRoleAssignment variable to true in naobehavior.cc. Note that this only enables the role assignment part of the algorithm and does not include the voting coordination system from the paper (which is not in the code base).
  
  You can specify position locations for the different roles in the getSpaceForRole() function in positioning.cc. The positioning system has two special roles: ON_BALL and GOALIE. The ON_BALL role is always assigned to the player closest to the ball while the GOALIE role is assigned to the goalie (the agent with uniform number 1) unless the goalie is the closest to the ball and has been assigned the ON_BALL role.
  
  A good way to get the current position assigned to an agent by the positioning system is to get the value returned by getSpaceForRole(worldModel->getProposedRoles()[worldModel->getUNum()-1]).
• Are there some additional possibly beneficial features I can turn on in the UT Austin Villa code base and experiment with? Yes.
  • Holonomic walk. You can use the holonomic walk described in the paper Using Dynamic Rewards to Learn a Fully Holonomic Bipedal Walk by setting the useHolonomicWalk variable to true in naobehavior.cc.
  • Faster walks. We have learned several walks that have top speeds a little over .8 m/s (the default walk has a top speed of .71 m/s). You can turn these on by uncommenting parameters for them in paramfiles/defaultParams.txt. This is done by removing the block comments (started by /* and ended by */) around one of the groups of walk parameters that begin with "target_fullparams", "balance_target2", or "balance_target3". Note that for each walk there are three sets of walk parameters (one each for going to a target, sprinting, and positioning).
  • Faster getup. We learned faster getup routines which can be turned on by uncommenting parameters for them in paramfiles/defaultParams.txt. This is done by removing the block comments (started by /* and ended by */) around "Faster getup off front params" for getting up when having fallen forward, and "Faster getup off back params" for getting up when having fallen backwards. If you don't see these parameter sets in paramfiles/defaultParams.txt you may need to update this file. Note that we didn't use the faster getup off front parameters in the competition as we felt they were a little unstable.
• Penalty Kick Shootouts: A penalty kick shootout occurs when two teams are tied after both regulation and overtime periods when a tie must be broken. Each time gets 5 attempts to score with a lone shooter against the other team's goalie. At the start of an attempt the ball is placed at the center of the field and the shooter has 60 seconds to score. The goalie for the defending team trying to stop the shooter is not allowed to leave the penalty box (the shooter is allowed to enter the penalty box). After the 5 scoring attempts by each team the team with the most goals wins. If the teams have the same number of goals after 5 attempts each then they continue alternating scoring attempts until one team scores and the other doesn't.
  • How do I create a penalty kick shooter? You can edit the methods in the PKShooterBehavior class in pkbehaviors.cc. In order to start a penalty kick shooter run the start_pkshooter.sh script.
  • How do I create a penalty kick goalie? You can edit the methods in the PKGoalieBehavior class in pkbehaviors.cc. You may find the isOutsidePenaltyBox() method useful which lets you know whether a location is inside the penalty box. In order to start a penalty kick goalie run the start_pkgoalie.sh script.
• I want to run repeated trials of an agent doing something and then measure X. Is there an automated way to do this? We have created agents for doing this sort of thing for reinforcement learning.
  • Look in the optimization directory. Here you will find scripting files for what we call optimization agents. Optimization agents perform some behavior (potentially determined by an input file) and then write out some output to a file. Check the README file for some pointers as to what is going on here.
  • Run the optimization/run-opt.sh script. This will by default run an agent that measures how far it can walk forward in 15 seconds.
    Example usage: run-opt.sh /tmp/out.txt
  • Checkout optimizationbehaviors.cc. Here you will find the behavior classes for the optimization agents. Methods for the classes are beam() to initialize where an agent starts, selectSkill() to control an agent's behavior, and updateFitness() to monitor and write out data.
• How can I play two teams against each other and automatically record the score from the game? We have a script for doing this which plays a single half between your team and an opponent team and then writes out the score.
  • Look in the stats directory. Here you will find scripting files for playing your team against an opponent team for a half. Check the README file for some pointers as to what is going on here.
  • Run the stats/run-stats.sh script. This will run your agent against an opponent agent, with your agent either starting on the left or right side of the field, and then write output to the given output file.
    Example usage: run-stats.sh /tmp/out.txt ~/opponent_dir left
  • Checkout recordstatsbehavior.cc. Here you can change the updateFitness() method in the RecordStatsBehavior class to change what data is being recorded and written out.
• Can I get an agent's ground truth position? Yes, you can use the worldModel->getMyPositionGroundTruth() method which gets an agent's ground truth position from the server. Note that this functionality will NOT be turned on during the class tournament so don't use this function in your final agent submission or you will be sorry. Also you will need to use a version of the server that has ground truth data enabled: rcroboviz_gt, run_server3d_gt, and rcsoccsersim3d_gt.
• Is there a way to move the ball or players around and also change the current game state? You can change the game state in RoboViz by typing 'O' and choosing a state from the menu. You can also select agents and the ball and then control-click to move them. Additionally you can use the Coach/Trainer command parser documented here to change the play mode as well as the positions and velocities of objects on the field. Examples of its usage can be found in move-ball.pl, stats/kickoff.pl, and optimization/beamball-and-kickoff.pl. The agent can also send commands directly to the Coach/Trainer command parser directly by passing the message to the setMonMessage() method. In order for this to work, however, the agent must connect to the server through the monitor port by adding the command line argument "--mport 3200".