RoboCup 2002 Rescue Simulation Semi-Final Rule and Evaluation

RoboCup 2002 Rescue Simulation Leagues


2002.2.13 draft(version.1)
2002.2.18 draft(version.2)
2002.3.11 draft(version.3)
caution: There may be some modification after test at Japan Competition (March 23-24) and computer condition on site.

Index

A) Schedules
  1. Team meeting: tentative(Jun. 18; from 13:00)
  2. Requirements;
  3. Preliminary games: (Jun. 19-21)
  4. Final games: (Jun. 22 - 23)
B) Game Rules
  1. Software Environments
  2. GIS Maps
  3. Agents
  4. Parameters used in games
  5. Evaluation rule (Which team will win ?)
  6. Valid games.
  7. Others

Schedules

  1. Team meeting: tentative(Jun. 18; from 13:00)
  2. Requirements;
    Each team is required to prepare following files:
    1. two gis_init data files,
      one for Kobe City map and the other for Virtual City map,
    2. galpolydata file and shindopolydata file.
    caution 1: Tools for these files can be download from Hatayama's tool and Takahashi's tool.
    The team can only change the number of ignition points, and the position of the ignition points agents.(cf. Game Rules [3].Agents)

    caution 2: The files that every team prepares are key for rescue operations. We hope that every team checks your own files by your agents.
    Thanks in advances for preparing good files.

  3. Preliminary games: (Jun. 19-21) The number of registered teams are fourteen teams at March.6,2002.
    1. They are divided in three groups A, B and C. Fourteen teams are labeled as {a1, a2, a3, a4, a5}, {b1, b2, b3, b4, b5} and {c1, c2, c3, c4}.
    2. At each league, all teams do games for Kobe City maps and Virtual City maps. caution 3: In this year, two kinds of map - Kobe City and Virtual City - are used. Each team will rescue at Kobe City map and Virtual City map with different situations.
       League P           |    Kobe City map       |  Virtual City map      |total 
                          +------------------------+------------------------+points
       gis_init data from | p1 | p2 | p3 | p4 | p5 | p1 | p2 | p3 | p4 | p5 |  
             =============+========================+========================+======
                       p1 |                        |                        |
                       p2 |    5*5 games           |    5*5 games           |   
                       p3 |     for A, B           |     for A, B           |
                       p4 |    4*4 games           |    4*4 games           |
                       p5 |     for C              |     for C              |
      
      comments:
      1. P is one of {A, B,C}, and x is the corresponding lower character.
      2. For games in league A (B, C), galpolydata and shindopolydata files are selected from files that a team of league B (C, A) designed.
    3. For example, the teams in the league A will do rescue operations at five times at Kobe city using different gis_int data files. The teams get points according to scores shown by LogViewers - the 1st team gets 4 points, the 2nd team gets 3 points, 3rd team gets 2 points, 4th team gets 1 points, and the 5th gets none. When the score of LogViewer are the same as other teams, then they get average points. And they do rescue operations at five time at Virtual City using different gis_int data files. The teams get points from 4*5*2=40 to 0 and are ranked by the points.
    4. Top two teams of each leagues are qualified to final games. When there are teams with the same points, then their total scores of LogViewer determine their ranks.
    Total (5*5 + 5*5 + 4*4)*2 = 132 games are done during 3days.
  4. Final games: (Jun. 22 - 23)
      Jun. 22
    1. Top 3 teams are selected in tournament style.
      The tournament games will be done between the 1st teams of a league (A1, B1, C1) and the 2nd teams of the other league (A2, B2, C2) - A1 vs. B2 , B1 vs. C2 and C1 vs. A2.
              A1 --+
                   +-->X: using data files of (KC_C1, VC_C1, CC_C2)
              B2 --+
              B1 --+
                   +-->Y: using data files of (KC_A1, VC_A1, CC_A2)
              C2 --+
              C1 --+
                   +-->Z: using data files of (KC_B1, VC_B1, CC_B2)
              A2 --+
      
      Each team will do two rescue operations, at Kobe city and Virtual City. The gis_init data, galpolydata and shindopolydata files are used from other league. For example, files of C1 are used at the game of A1 vs. B2. In a case of tie score, third game will do in the follwoing steps:
      • select map by coin toss, head -> Kobe city map, tail -> Virtual city map,
      • use data files of C2.
    2. exhibition of Fukuoka map by three teams (X_loser, Y_loser, Z_loser).
    3. semi-final matches among (X_winner, Y_winner, Z_winner).
      Top two teams are finalist (F1, F2). The third team is the third-place winner (F3). Each team will do three rescue operations, using gis_init data, galpolydata and shindopolydata files of X_loser, Y_loser, Z_loser. A coin toss determines which map, Kobe city or Virtual city is used. (head -> Kobe city map, tail -> Virtual city map).
              map:  coin toss by |X_winner|Y_winner|Z_winner|
              -------------------+--------+--------+--------+total
                   data files of |X_loser |Y_loser |Z_loser |points
              ===================+==========================+======
                         X_winner|        |        |        |
                         Y_winner|        |        |        |
                         Z_winner|        |        |        |
      
      The teams get points according to scores shown by LogViewers - the 1st team gets 2 points, the 2nd team gets 1 point, 3rd team gets none. When the score of LogViewer are the same as other teams, then they get average points. The teams are ranked as F1, F2, F3 by the total points of three rescue operations. When there are teams with tie scores, the teams do rescue operations.
      	(extra 1)      map : the other map of X_winner's coin toss
                        data file: X_loser
      		             If the ties are resolved, then finish,
                                                              else do extra 2.
      	(extra 2)      map : the other map of Y_winner's coin toss
                        data file: Y_loser
      		             If the ties are resolved, then finish,
                                                              else do extra 2.
      	(extra 3)      map : the other map of Z_winner's coin toss
                        data file: Z_loser
      
    4. Jun. 23
      Final games between F1(1st team in the semi-final) and F2(2nd team).
           (1st game) using      map: coin toss by referee.
      	             data file: committee provides.
                      If the winner of the 1st game is F1, 
                         then F1 is the champion of 2002, 
      		   else do 2nd game.
           (2nd game) using      map: the other map used in the 1st game.
      	             data file: committee provides.
      		The winner of the 2nd game is the champion of 2002.
      

    Game Rules

    1. Software environments:
      A Linux package will be used in Fukuoka. The package will be down loaded here . On Feb. 13, 2002, a package (version number 0.37) including newly proposed softwares can be downloaded. According voting to proposals, this year we will use newly
      1. traffic simulator proposed by Morimoto,
      2. civilian agent with behavior rule sets proposed byShinoda and Noda [tentative, cf ].
    2. GIS Maps:
      This year, two kinds of GIS map are used for games,
      1. 1/10 model of Kobe map,
      2. 1/1 model of virtual city map.
      Each team must prepare gisinit.txt (that determines initial condiations of agents) for both maps. The files are submited at Team meeting at Jun.
      The files can be edited by using Hatayama's tool .
    3. Agents:
      1. The numbers of agents adn ignition points specified in gisinit.txt are in the following range.
                                    min.  max.
               -------------------+-----+-----
               Fire Brigade       :  10   15 
               Police Force       :  10   15 
               Ambulance          :   5    8 
               Civilian           :  70   90 
               Fire Brigade Center:   0    1
               Police Force Center:   0    1
               Ambulance Center   :   0    1
               Refuges            :   1    5
               -------------------+-----+-----
               ignition points    :   2    8   
               
      2. agents prepared by committee is only Civilian:
        Several candidates of civilian agent have been proposed.
        caution 4:
        Committee members think that Shinoda and Noda's civilian with behavior rule sets is nice for us to try. The problem is that it does not work stable at present. Committee members decide the following procedeure:
        1. at RoboCup Japan Open (March, 23-24, 2002), Shinoda and Noda's civilian with behavior rule sets will be tested,
        2. if there is no problem,
          then use Shinoda and Noda's civilian with behavior rule sets,
          else use Koto's sample civilian.
      3. agents that a team prepares:
        A team is recommended to make folloing all kind of agents.
        • Fire Brigade:
        • Police Force, Ambulance:
        • Fire Brigade Center, Police Force Center, Ambulance Center:
        However, a team can participate as long as
        1. the team must make original Fire Brigade agent by itself.
        2. the team uses Police Force, Ambulance agents that are open in the net, and express acknowledgements to their usage.
        3. the team can Not use other team's Fire Brigade Center, Police Force Center, and Ambulance Center. Because, communication between **-Center agent and corresponding agents is one of key reseach themes in multi-agent system.
        When
        team A := makes only Fire Brigades,
        team B := makes Fire Brigades, Police Force.
        team C := makes Fire Brigades, Police Force, Ambulance.
        team D := makes Fire Brigades, Police Force, PF(Police Force) Center.
        , then their rescue teams are compose of the followings.
        * own agents agents from OPEN source
        Team A Fire Brigades Police Force, Ambulance
        Team B Fire Brigades, Police Force Ambulance
        Team C Fire Brigades, Police Force, Ambulance  
        Team D Fire Brigades, Police Force, PF Center Ambulance
    4. Parameters uses in games:
      1. new parameters in config.txt[kernel]:
        caution 5: (many of these are referef to "Fir Play rules" last year, and they are checked by kernel from 0-34 version.)
        Committee members decide the following procedeure:
        1. in principle, use Kernel with default config.txt,
        2. the parameters in default config.txt will be tested at RoboCup Japan Open (March, 23-24, 2002),
        3. if there is no problem,
          then use default config.txt,
          else specify parameters in config.txt as rules.
        The default parameters are
        $ pwd
        /../rescue-0_37-unix/RUN
        $ more config.txt
        
        ################################################# new options for the 2002 competiti\on.
        
        area_per_repair_cost            : 20000000
        round_down_quantity             : false
        accept_multiple_nozzles         : true
        near_agents_rescuable           : true
        steps_far_fire_invisible        : 5
        steps_agents_freezed            : 3
        notify_initial_position         : false
        notify_position_history         : false
        miscsimulator_supports_load     : false
        notify_only_fire_for_far_buildings      : true
        
        notify_unchangeable_informaion          : true
        #                       true for compatibility
        ################################################# traditional options
          :
             
        For more infromation, please refer to Koto_san's mail with Subject: package ver. 0.34 available.
      2. parameters: [tentaive]
        1. simulation time: 300 steps,
        2. range of eyesight is 10m,
        3. range of voice is 30m,
        4. power of earthquake.
          caution 6: Committe will prepare several files - shindopolydata.dat, galpolydata.dat - that determines how much roads and building are collapsed. Any pair of the files will be used for games. So it is no use to program the information of these files, BUT, it is PROHIBITED to program the information of these files - gisinit.txt, shindopolydata.dat, galpolydata.dat - directly.
        5. number of communication:
          all agents can tell (or say ) Max 4 sentences (in terms of simple sentence) per 1 turn.
          Agents in the same rescue team can hear other team mates saying.
          The agent can see sender ID of all messages, but can read MAX 4 messages. (The idea is that the agent can select the best four messages from ones it received.)
          There is no restriction on the length of messages, however, the length under 256 byes is recommended from network traffic.
    5. Evaluation Rule(Which team will win ?):
      very important caution 7: (this term is differnt from last year.) 2002's evaluation forumla is designed to reflect activities of all kinds of agents.
          V=(P + S/Sint) * sqrt(B/Bint)
      
            P: number of living civilian agent,
            S: remaing HP of all agents,
            Sint: total HP of all agents at initial,
            B: area of houses that are not burnt,
            Bint: total area at initial.
      
      The more V value for a map, the better rescue operation.
    6. Valid games:
      Are games valid when some simulators fail?:
      Some simulators may fail during a game, although they are debugged. In that case, only ONE more simulation is permitted. When the same problem occurs at the simulation, the simulation continues to the end.
      The point at the game is treated on the principle that the continuation should not be the team's favor. Namely,
      1. in a case of the traffic simulator fails, the continuation is against the team's point. So, the game is valid and the point is treated as effective one.
      2. in cases of the other simulators (including kernel) fail, the continuation is for the team's point. For example, the fire simulator stops then there are not more fire. In that case, the team gets 0 point for that simulation in the preliminary games, and loses the game in the final games.
    7. Others
      1. We have to ensure that the competition field for two competing teams is identical not only parameters but also the seeds of random numbers used in simulators, -- the miscsimulator, collapse simulator, fire simulator.
        caution 8: This does NOT say that the same seeds and parameters are used for all games. The committe will change them in codes, files - galpolydata.dat, shindopolydata.dat - properly.
      2. remote entry is OK,
      3. source code must be open after competition,
      4. Log files will be open.