next up previous
Next: Discussion and Related Work Up: Results Previous: Team Coordination

RoboCup-97

The RoboCup-97 simulator competition was the first formal simulated robotic soccer competition. With 29 teams entering from all around the world, it was a very successful tournament.

It was in preparation for this competition that the team member agent architecture, including both the flexible teamwork structure and the inter-agent communication paradigm, was developed. Since competitions are not controlled experiments, their results are not presented as scientific validation of our individual techniques. Such validation is presented in Section 7.1 and 7.2. However, qualitative evaluations and anecdotes from these competitions can provide insights into the strengths and weaknesses of various approaches.

Table 8 shows the results of CMUnited-97's games in this tournament. CMUnited-97 won 3 of its first 4 matches by wide margins, with the other match being a close victory. Its 5th opponent, FCMellon, was also our own team and was identical to CMUnited except that it did not use a flexible teamwork structure: players did not switch positions, did not use flexible positioning of any sort, and did not use set-plays. Our goal in entering FCMellon in the competition was to highlight the impact of our research contributions in CMUnited. Due to the results reported in Section 7.1, we expected CMUnited to win this game. Before the game between CMUnited and FCMellon, FCMellon won its 4 games by a combined score of 49-4.

   table467
Table 8: The scores of CMUnited-97's games in the simulator league of RoboCup-97. CMUnited-97 won 5 of its 7 games, finishing in 4th place out of 29 teams. *The last game was lost by one goal in overtime.

The subsequent game was against the eventual tournament champion AT-Humboldt [5]. As described in Section 7.1, CMUnited-97 used a 4-4-2 formation in general, switching to an 8-2-0 formation if winning near the end of the game, or a 3-3-4 formation if losing. The triggers for these formation switches were defined as part of the locker-room agreement. However, by the time CMUnited-97 played against them, it was clear from watching other games that AT-Humboldt was the team to beat. Therefore, we altered the team's locker-room agreement so that it would adopt a more conservative, defensive strategy at the beginning of the game. As a result, AT-Humboldt scored fewer goals against CMUnited-97 than it did against any of its other competitors. The 6-0 result might have been even closer had CMUnited-97 not switched to the more offensive 3-3-4 formation near the end of the game when it was losing in an attempt, though unsuccessful, to score some goals.

One of the main advantages of the AT-Humboldt team was its ability to kick the ball harder than any other team. Its players did so by kicking the ball around themselves, continually increasing its velocity so that it ended up moving towards the goal faster than was imagined possible. Since the soccer server did not enforce a maximum ball speed, a property that was changed immediately after the competition, the ball could move arbitrarily fast, making it impossible to stop. With this advantage at the low-level behavior level, no team, regardless of how strategically sophisticated, was able to defeat AT-Humboldt.

Having lost in the semi-finals, CMUnited-97 then played in the 3rd-place game against ISIS [40]. CMUnited-97 scored first in this game off of a corner-kick set-play. However, ISIS equalized near the end of the game and the game went to overtime. ISIS scored to win in what proved to be one of the more exciting matches of the tournament.



next up previous
Next: Discussion and Related Work Up: Results Previous: Team Coordination



Peter Stone
Thu Dec 17 15:26:44 EST 1998