Strategy for Robo-Kabaddi

Size: px

Start display at page:

Download "Strategy for Robo-Kabaddi"

Nickolas Fisher
5 years ago
Views:

1 Strategy for Robo-Kabaddi

2 Two Strategies are required One for raiding the opponent. Another, while the opponent is raiding us. Aggressive/Defensive Stance

3 Initial Configuration By intuition & with out loss of symmetry. All members are located on arc of a circle (or ellipse) at equal distances. The distance between the deepest point of the arc and the gate-line is ⅔ of the length of the court.

4 When we are raided by the opponent Attacking When majority of the team members are active. Only when there are more than 2 agents. Defending When very few are active (may be 1 or 2) Its easier to raid than being raided

5 Risk Parameter The court is seen as a matrix of size mxn. Each cell has an attribute called risk value used to make sure that the anti-raider is safe enough to attack the raider. It is updated as the game progresses.

6 Initial distribution of the risk value The order of riskiness is 1 < 2 < 3 < 4 < 5 < 6 i.e. grids with label 1 are the least riskiest locations possible.

7 Factors for updating the risk value If the raider was blocked successfully earlier in a particular grid then its risk value is decreased and vice-versa. Time spent in a particular grid. Risk increases as time increases. All the grids it has traversed successfully. (i.e. if the raider goes back with out touching an anti-raider).

8 Aggressive Stance

9 Issues to be dealt Decision on the number of agents to use to block the raider Path Planning path taken to block the raider Surround the raider dynamically!! Using the agents to block the path taken by the raider to escape

10 Path Planning Red/Green - Trapping Distance Blue Shortest Distance Decision on the curviness of the path is based on various parameters Trap and Escape Velocities Shortest distance between the raider and gate line Trapping distance Could be elliptic or parabolic or any other conic section path.

11 Guessing the raiders move before hand We keep track of the n recent positions of the raider taken at regular intervals (say t). We need to guess this as accurately as possible at least when it is about to attack (touch). (Currently we take the lenience that our guess can be wrong when its time passing or is near gate-line). Since it least probable that the raider will follow a particular curve, we use the recent 4-5 positions of the raider and do curve fitting.

12 Defensive Stance Pass the time by moving here & there away from the raider. Moving deep in the court. When very few are active (may be 1 or 2) Its easier to raid than being raided

13 When there are 2 anti-raiders We move them at the two corners of the court (i.e. to the cells with lowest risk value). When the raider approaches one anti-raider, the other one starts surrounding the raider by maintaining enough shortest distance. This is just to scare the raider and not to attack it as containing the raider with only two antiraiders is difficult.

14 When there is only 1 anti-raider We place it in one of the grids with low risk value and make it move in concentric circles (with varying the radius in its range)

15 While raiding the opponent Aggressive Mode Your suggestions are required. Very Aggressive Mode When the number of anti-raiders are very less, say 1 or 2, it is easy for the raider to make the antiraiders inactive. (We may think to make a compulsory constraint to get at-least one of the anti-raider out)

16 Aggressive Mode It is more likely that the outer most antiraiders try to surround. A similar risk map is developed which learns automatically as the game progresses. Touch the anti-raider which is in a cell with risk value less than a threshold. Path Planning to escape RRT can be one option

17 Probability Model Interest is in finding P( raider to cross the gate line successfully ) Useful for Energy Management P( raider r to cross the gate line successfully ) 1 - normalized distance between the gate line and the raider No. of anti-raiders behind it with shortest distance less than certain threshold Width-Gap between itself and boundaries

MULTI AGENT GAME OF KABADDI: STRATEGIES AND SIMULATION

MULTI AGENT GAME OF KABADDI: STRATEGIES AND SIMULATION Thesis submitted in partial fulfillment of the requirements for the degree of Master of Science(by Research) in Computer Science and Engineering by