Contest: Pacman Capture the Flag

Enough of defense,
Onto enemy terrain.
Capture all their food!

[5/5] Final tournament rules, brackets and maps announced

[4/13] New information about submitting to the nightly tournament has been added to the contest details section at the bottom. The nightly tournament is how you qualify for the finals.

Final Tournament

The final tournament will be held live in Soda 273 at 5pm on Wednesday May 6th. Bring your agents with you!

Each match-up will be decided in three games. The first game will be played on defaultCapture.lay. The other two games will be played on one of the 20 randomly generated maps now included in this distribution as layouts/contest*Capture.lay. One of these mazes can be chosen at random using mazeGenerator.py.

To watch games from the last nightly tournament, download the server logs and use the unpack.py tool. python unpack.py [server-log] [team-name] will generate replay files, which can be replayed with python capture.py --replay [replay-file]. The server log for May 5 is corrupted, unfortunately.

The winner's bracket:

Introduction

The course contest involves a multi-player capture-the-flag variant of Pacman, where agents control both Pacman and ghosts in coordinated team-based strategies. Your team will try to eat the food on the far side of the map, while defending the food on your home side. The contest code is available as a zip archive.

Academic Dishonesty: While we won't grade contests, we still expect you not to falsely represent your work. Please don't let us down.

Rules of Pacman Capture the Flag

Scoring: When a Pacman eats a food dot, the food is permanently removed and one point is scored for that Pacman's team. Red team scores are positive, while Blue team scores are negative.

Eating Pacman: When a Pacman is eaten by an opposing ghost, the Pacman returns to its starting position (as a ghost). No points are awarded for eating an opponent. Ghosts can never be eaten.

Winning: A game ends when either one team eats all but two of the opponents' dots. Games are also limited to 3000 agent moves. If this move limit is reached, whichever team has eaten the most food wins.

Computation Time: In online play, each agent will have only 0.5 seconds to choose an action. If an action is not returned in this amount of time, the server will move the agent randomly.

Observations: Agents can only observe an opponent's configuration (position and direction) if they or their teammate is within 5 squares (Manhattan distance). In addition, an agent always gets a noisy distance reading for each agent on the board, which can be used to approximately locate unobserved opponents.

Getting Started

python capture.py

A wealth of options are available to you:

python capture.py --help

python capture.py -b OffenseDefenseAgents 

Online Games

python pacclient.py -a OffenseDefenseAgents

python pacclient.py -U blitz -P noprisoners -a AllOffenseAgents

Any agent that works in a local game should work equivalently in an online game. However, there is a real-time element to the game: you have only 0.5 seconds (plus a little leeway for network lag) to choose a move. The server will enforce this time limit, and will choose a move for you if you don't supply one in time.

Designing Agents

Interface: The GameState in capture.py should look familiar, but contains new methods like getRedFood, which gets a grid of food on the red side (note that the grid is the size of the board, but is only true for cells on the red side with food). Also, note that you can list a team's indices with getRedTeamIndices, or test membership with isOnRedTeam.

Finally, you can access the list of noisy distance observations via getAgentDistances. These distances are within 6 of the truth, and the noise is chosen uniformly at random from the range [-6, 6] (e.g., if the true distance is 6, then each of {0, 1, ..., 12} is chosen with probability 1/13). You can get the likelihood of a noisy reading using getDistanceProb.

To get started designing your own agent, we recommend subclassing the CaptureAgent class. This provides access to several convenience methods. Some useful methods are:

  def getFood(self, gameState):
    """
    Returns the food you're meant to eat. This is in the form
    of a matrix where m[x][y]=true if there is food you can
    eat (based on your team) in that square.
    """

  def getFoodYouAreDefending(self, gameState):
    """
    Returns the food you're meant to protect (i.e., that your
    opponent is supposed to eat). This is in the form of a
    matrix where m[x][y]=true if there is food at (x,y) that
    your opponent can eat.
    """
    
  def getOpponents(self, gameState):
    """
    Returns agent indices of your opponents. This is the list
    of the numbers of the agents (e.g., red might be "1,3,5")
    """
    
  def getTeam(self, gameState):
    """
    Returns agent indices of your team. This is the list of
    the numbers of the agents (e.g., red might be "1,3,5")
    """
    
  def getScore(self, gameState):
    """
    Returns how much you are beating the other team by in the
    form of a number that is the difference between your score
    and the opponents score. This number is negative if you're
    losing.
    """
    
  def getMazeDistance(self, pos1, pos2):
    """
    Returns the distance between two points; this is either
    the Manhattan distance early in the game, or actual
    shortest path maze distances once the computation is
    complete.
    
    The distancer computes the shortest path between pairs of
    points in the background, and starts using them as soon as
    they are ready. These are not just pre-computed ahead of
    time because of the time limit - we don't want to lose our
    turn because we're computing distances we don't need yet!
    """

  def getPreviousObservation(self):
    """
    Returns the GameState object corresponding to the last
    state this agent saw (the observed state of the game last
    time this agent moved - this may not include all of your
    opponent's agent locations exactly).
    """

  def getCurrentObservation(self):
    """
    Returns the GameState object corresponding this agent's
    current observation (the observed state of the game - this
    may not include all of your opponent's agent locations
    exactly).
    """
  

Baseline Agents: To kickstart your agent design, we have provided you with two baseline agents. They are both quite bad. The OffensiveReflexAgent moves toward the closest food on the opposing side. The DefensiveReflexAgent wanders around on its own side and tries to chase down invaders it happens to see.

To facilitate agent development, we provide code in distanceCalculator.py to supply shortest path maze distances as soon as they can be computed, but to supply Manhattan distances until then. This approach demonstrates some of the techniques you might want to use when designing agents under time constraints (e.g., sharing information and computing in parallel). However, this is neither the only nor the best way to solve the maze distance problem.

Restrictions: You are free to design any agent you want, and you need not use any of the code we have provided. Because the tournament will be run using the online architecture, you can run your agent from your machine using any resources or code you want; you can even write code in different programming languages if you so desire. Change pacclient.py at your own risk, though, because you don't want to break the network functionality.

Loading agents: Agents are chosen by agent factories, also specified in captureAgents.py. You may also pass options to your agent factories (see --help), which can then be passed on to your agents.

Contest Details

• Qualifying: Every night, we will post the results of a double-elimination tournament among all submitted agents, including two sets of "qualifying agents". To qualify for the final tournament, you must first submit your agent using these instructions. Then, you must be ranked ahead of at least one of the qualifying agents in one of these nightly tournaments before the qualification deadline. The earlier you submit your agents, the more chances you have to qualify!
• Exhibition: The current best agents by nightly ranking from Berkeley and Utah will compete for the glory and honor of their respective universities.
• Tournament: A final double-elimination tournament will be conducted live in the basement of Soda hall on the evening before the last day of class (Wednesday 5/6). The final lecture the next day will include replays of important matches. The final tournament will be similar to the defaultCapture layout.

Important dates (subject to change):

Teams: You may work in teams of up to 5 people.

Prizes: The top three teams will receive awards in class on Thursday 5/7, including shiny medals and extra credit points.

• First Place: 1% overall point increase
• Second Place: 0.75% overall point increase
• Third Place: 0.5% overall point increase

Bonus prizes: If at least 30% of the class qualifies for the tournament, then all prizes will be doubled. Additionally, all qualifiers will get a 1% bonus on their final exam. Note that 2% overall (the top prize) is a lot: it is equivalent to 6.7% on the final.

Named Games

python pacclient.py -a KeyboardAgents -g MyCoolGame

This being an inherently social feature, the CSUA has set up an IRC channel for chatting about the project. It can be reached at:

Server: irc.csua.berkeley.edu
Channel: #cs188

PyGame Graphics

Once you have installed PyGame, you can run either the networked or local game using the improved (speedier) graphics.

python capture.py -1 RandomAgent -G

Acknowledgements

Have fun! Please bring our attention to any problems you discover.