Ut2013 learning notes

Challenge description

A.SPL Challenge

SPL The challenge requires every team to have players , If there are six teams participating , Each team contributes One or two substitute players . If you use two players from the same team , They are playing on the same side . Both teams are composed of randomly selected substitutes , Each contestant participates 4 A substitute game , Every game lasts 5 minute . Shorter games are used to allow more games to be played within a specified time . In the normal SPL In the game , The goalkeeper is designated as a goalkeeper at the beginning of the game . In the challenge , The first defender to enter the penalty area became the goalkeeper of the game .

During the challenge , Players are allowed to communicate with each other using a simple protocol , But this kind of communication is unnecessary . The agreement allows communication of the player and the position of the ball , Variance of player and ball position （ uncertainty ）, The speed of the ball , Time since the last time I saw the ball , And whether the robot falls or is punished .

SPL The challenge uses two indicators to score ： Average goal difference and average score of three referees . These two scoring indicators are combined to determine SPL Challenge champion . Human referees are used to help Identify good teamwork ability in agents , And in a limited number of games Reduce the influence of random variance . For every game , Every referee is asked to score every player , Score at 0（ Bad ） To 10（ good ） Between . The judges were asked to focus on the ability of teamwork , Not personal skills , Such a robot with low skills can still get high scores for good teamwork .

B. Two dimensional challenge

about 2D Substitute challenge , Every team has contributed Two substitute players , Two of the teams are composed of substitutes . The competition includes two 5 Half a minute , Team has 7 player , Not the standard one 11 player . The number of players has been changed , Because most teams release based on the same code , It's for 11 Player teams provide default formation , Provide implicit coordination for these teams . The seventh player of each team is from agent2d Our goalkeeper , Coach is not used in the challenge . Substitute players are encouraged to use default agents 2d Communication protocol for communication , But it's not necessary .

Players pass Average goal difference score ; There is no human referee . In order to accurately measure their performance , Each team plays at least one game with each other's opponents . A total of nine teams participated in this challenge . Game pairing is a Greedy algorithm Select the , The algorithm tries to balance the number of cooperation and confrontation between agents from different teams , Such as Alg．１ Shown . The algorithm is universal , It can be applied to other special team cooperation settings . When all agent With other teams At least one game when , The algorithm stops .

C.3D Challenge

about 3D Substitute challenge , Each team contributed to a match Two substitute players . The game lasted for two 5 Half a minute , Both teams are made up of substitute players . In the challenge There is no goalkeeper To increase the possibility of scoring .

Every substitute can communicate with his teammates about the ball and his position , But using this protocol is optional . The challenge score is only obtained by the agent in all competitions Average goal difference decision . During the challenge, there were four temporary matches ,10 All teams participated in every competition . The team matching of the competition is made by Alg.１ decision . So every substitute player plays at least one game .

 

Greedy algorithm gets Locally optimal solution , Select the conditions that should meet the local optimization

Dynamic programming is to find the global optimal solution

The basic idea ： 1. Build a mathematical model to describe the problem

2. Divide the problem into several sub problems

3. Solve every subproblem , Get the local optimal solution of the subproblem

4. The local optimal solution corresponding to the subproblem is synthesized into a near of the original whole problem    Quasi optimal solution

Analysis of competition results

* SPL（ Standard group ） match

  * drop-in challenges The team with good performance in the team has strong cooperation ability , But not necessarily the best low-level skills

  * On the whole , Race results and drop-in challenge Is positively correlated

* 2D match

  * The race data sample is not enough , Later, the teams played against UT The standard team matches are 1000 site , The average value and standard deviation of goal difference are recorded

  * drop-in challenge There is no significant direct relationship with the race results , But in the standard team competition （ Zheng sai ） Teams that do well in tend to do well in drop-in challenge Better performance in

  * drop-in challenge An important link is adaptation （ From other teams ） Teammate

  * Dynamic role assignment can be improved in drop-in challenge Performance in

* 3D match

  * drop-in challenge There is no significant direct relationship with the race results , But in drop-in challenge The team that performs well in the team tends to compete in the standard team （ Zheng sai ） Better performance in

* Drop-in Challenge Strategic analysis

  * Build four teams to run and V-C The same as described in part drop-in match

    * Dribble: Never play football , With ball only

    * DynamicRoles: Dynamic role assignment

    * NoKickoff: Do not pass to the side of the ball to kick off ？

    * UTAustinVilla: The program used in the competition

    * Dribble > UTAV > NoKickoff > DynamicRoles

  * Conclusion

    * When no teammate kicks off, passing it to the side of the ball to kick off will help improve the performance （UTAV > NoKickoff）

    * play football / Passing will negatively affect the result . Probably because UT Low level skills are better than drop-in Teammate , So passing the ball to them will reduce the level of the team