Home

Welcome to the Simurosot wiki!

#Summer 2015#

1. Making a new strategy for simurosot-
   • Spin was implemented in the Goalkeeper, new prediction function to position the goalkeeper.
   • Cover Goal was synchronised with the goalkeeper to reduce chances of own goals. Charge was also implemented in the CoverGoal when the area around the goal is clear.
   • Midfielder made to defend when the ball is on our side and attack when the ball is in the opponent half. It blocks the opponent bot closest to ball while defending.
   • A few changes in the prediction of the position of the ball were made. In tAttack, shooting moving ball is implemented.
   • Recieve Ball implemented to attack the ball when the attacker looses the ball else position itself in the reflecting position.
   • A single Play made with one of these tactics given to each bot.
2. Q-learning (SARSA) added to Simurost. It has been implemented on simurosot. Training has also been done. It was working as expected. (Q-learning explained in attached document.)

##Research Papers##

• A New Passing Strategy Based on Q-Learning Algorithm in RoboCup

  This paper discusses how the team TiJi-GDUT2008, developed a new algorithm for a more effective passing for Robocup SSL. Passing is very different from play-switching (and has diff. set of algorithms altogether) just because it involves multiple agents. So, many previous Q-Learning algorithms were named, like - Minimax (only effective for zero-sum),FOF, Nash-Q and some advanced ones like GIGA,GIGA-Wolf.

  Xiong, Li, et al. "A new passing strategy based on Q-learning algorithm in RoboCup." Computer Science and Software Engineering, 2008 International Conference on. Vol. 1. IEEE, 2008.

• Finding the best pass-receiving position in RoboCup

  The paper starts off with describing how other teams used to implement passing. CMU earlier used to divide the field, and followed a fitness-evaluation thing to find the best position. One used to calculate a function based on various parameters such as bot-ball distance, number of opponents near to the ball etc. The author also introduced the concept of an ‘open-angle’ into the criteria. He discussed two approaches. Explicit enumeration : He used a fitness function with various weights for various parameters ( home, away bot-ball distance, open-angle) and then applied some constraints (max-bot-acc. , max-bot-speed etc, our bot must reach before opponent). Downside is, it can take upto 30 sec to get to the solution. Meta-heurisitc approach: choose-and-check approach. He used a ‘CuckooSearch’ algorithm for finding that position, using Levy flights and all. Yoon, M., and T. Lane-Visser. "Finding the best pass-receiving position in the RoboCup Small-Size League." 41st Annual Conference of the Operations Research Society of South Africa. 2012.

• Learning to Role-Switch in Multi-Robot Systems

  This paper discusses an approach to coordinate a team of robots or implement effective role-switching though learning. What they did is reduced the state space from infinite states, to a finite number of states, by considering possession, ball position, bot-ball distance etc. Then a DP or recursion (in whole, iteration based ) based Q-learning algorithm is applied.

  Martinson, Eric, and Ronald C. Arkin. "Learning to role-switch in multi-robot systems." Robotics and Automation, 2003. Proceedings. ICRA'03. IEEE International Conference on. Vol. 2. IEEE, 2003.

• Dynamic Model and Shooting Algorithm on Simurosot

  A point on the goal is found according to the opponent GoalKeeper position and velocity where shooting will give most chances of scoring a goal. After the goal point found we find the required velocity and angle to hit the ball so the final velocity of ball is along the direction of line joining the ball and the target goal point. We go along elliptical path to shoot the ball.

  TW, Y., Chan-Tan, Y. W., Lee, H. A., EL, C. T., Jiang, H., & Sng, H. L. Dynamic Model and Shooting Algorithm on Simurosot.

• Defence Strategy using Safety Region Method

  This paper basically analyzes the positions of our defensive robots and defines a safety region . This is the region from where if the opponent bot shoots, our defence or keeper is able to stop the ball from reaching the goal . Moreover there is an algorithm to calculate the approximate safety area in a very fast computation time and hence it can be calculated every frame .This paper talks about changing our defensive strategy, no of defensive bots and their positions based on this safety region . Hence the defensive bots will move in a position such that the position around the ball becomes a safety region.

  Inagaki, Taro, et al. "Robust algorithm for safety region computation and its application to defense strategy for RoboCup SSL." RoboCup 2011: Robot Soccer World Cup XV. Springer Berlin Heidelberg, 2012. 484-494.

• Study on Shooting Skill in Robocup SSL

  It decides the best position to shoot at opponent’s goal . It takes in factor of opponent bot inside the opposition goal , it calculates the angle formed between the vectors joining the ball and the two obstacle points . It shoots through the maximum angle available .If that maximum Angle is less than a particular threshold it dribbles forward until the threshold is reached. Yang, Zeng-Guang, et al. "Study on shooting skill in RoboCup Simulator League." Machine Learning and Cybernetics, 2003 International Conference on. Vol. 4.IEEE, 2003.

• Decision making for a Robocup multi-agent system

  It talks about the different tactics to be included and what works it should execute in what order.

  Vetulani, Zygmunt. "Decision making for a robocup multi-agent system." Robot Motion and Control, 2002.RoMoCo'02.Proceedings of the Third International Workshop on.IEEE, 2002.

• STP: Skills, Tactics and Plays for Multi-Robot Control in Adversarial Environments

  Describes about the execution of tactics used by CMDragons and the pass and receive States.