IAS beta1.0 User Manual
By Tarek Hassan
Introduction
IAS = In the Agent’s Shoes
This tool was developed to put a human user in the place of a Robocup agent for the purpose of logging their actions and their reactions to different game situations. This log could then be used for machine learning for development of software agents. IAS could also be used to train, test or challenge an AI team by allowing the human players to create desired game situations through their actions (e.g. try different attack approaches to challenge a goalie agent’s adaptability).
Display
Your player’s view is visualized in a bird’s-eye perspective (top-down). Everything in the display is relative to your on-screen player (i.e. the player and their view cone are static and everything else moves relative to them). In this version of IAS the user is restricted to seeing only what the agent can see. Additional visual information that the agent can’t ‘see’ (like field lines that aren’t directly in-front of the player)could be deduced and added, but that would depend on and affect the actual implementation of the learning process. This also applies to the addition of more advanced controls(see Controls below).
This is what the IAS display looks like.
For more information about any of the objects in the display see the SoccerServer manual.
Controls
Dashing:
Clicking the left mouse button will make your player dash forwards with a power relative to the distance of the mouse click from your player on screen.
Holding down the left mouse button will make your player repeat that same dash every cycle until the button is released or another action is performed.
Turning:
Clicking the right mouse button will make your player turn with a moment equal to the angle of the mouse click from the imaginary line straight ahead (upwards) of your player.
Kicking:
Pressing the space-bar will make your player kick the ball with a power relative to the distance of the mouse pointer from your player(like dash) and an angle equal to the angle of the mouse pointer (like turn).
Again, more advanced controls (such as dribble) would be easy to add to the InputHandler.java class as combinations of these basic commands, but their addition would have to be also incorporated into the learning model.
More to come….