Design of Emergent and Adaptive Virtual Players in a War RTS Game - Foundations on Natural and Artificial Computation - page 374

in VR u ) agent u has to move in direction to the region where there are more
enemies according to the global information of the army.
- Group/Run away: the soldier has to group with other team mates placed
inside its visual range; if there are not mates in this range, then the agent
has to move towards the position where its flag is placed.
- Move forward objective: the agent advances towards the opponent's flag
position if this is known, otherwise it moves randomly.
- No operation: execute no operation.
- Explore: the soldier moves towards a non-explored-yet region of the map.
- Protect flag: The soldier has to move towards the position of its own flag; if
it is already near to it, unit just should scout the zone.

Perceptions . Each unit knows its specific situation in the game from the percep-
tions that receives in its visual range. In our game, the agent state is determined
by its health (measured in three range values: low, medium and high), and the
following boolean perceptions:

1. Advantage state? ( S a ): “the unit is into a handicap situation”. This percep-
tion is true if, according to the information in its visual range, the number
of mate soldiers is higher than the number of rival units, and false otherwise.
2. Under attack? ( U a ): this perception is true of the soldier is involved in a full
body combat with another opponent unit, and false otherwise.
3. Objective visible? ( O v ): the opponent's flag is visible in its visual range.

Game objective . The army that first captures the rival flag is the winner; if this
situation is never reached after a number of turns, then the winner is the player
whose army inflicts a higher damage to its rival army and where the damage is
measured as number of dead units.

2.1 Notes on Specific Issues

Observe that 'fighting' is not a specific action to be executed by the army units.
This should not be surprising because 'fighting' is not usually controlled by
players in standard RTS games as this action should often is executed at the
unit level and not at the army level; in our game this action is automatically
executed when a unit and a rival soldier meet in any grid of the scenario.
Also, note that five of the six possible actions that units can execute require
to make movements in the game world. This basically means that some kind
of pathfinding should be processed to do the movement realistic. Classical al-
gorithms such as A ∗ were not good candidates for this task since, as already
indicated, most of the scenario is not known for the army units. A practical so-
lution was to let units make a good-enough movement depending on its scenario
knowledge; several cases were covered: Concave obstacles : the unit moves by
considering different angles with respect to its position in the scenario; Convex
obstacles : the unit moves by following the obstacle contour in a specific direction;
and Bridges and narrow passages : a simple ant-colony algorithm was executed.