public AIAgent(int player) : base(player)
{
playerNum = player;
opponentNum = playerNum == 1 ? 2 : 1;
// Create a new, default QLearner. Usually want to replace this value for more control
QLearner = new QLearner();
IsLearning = false;
// Set strategy
strategy = Strategy.StrategyWithType(playerNum, StrategyType.RunAway);
level1Searcher = new DiscreteAdversarialSearch(playerNum,
strategy.Level1Heuristic,
getFillerAction,
getNewPathIndex,
Level1StepSize,
4);
decisionTimer = 0;
level2Searcher = new AStar(Level2MaxNodesInPrioQueue, Level2MaxExpansions, strategy.Level2CostFunction,
strategy.Level2GoalFunction, strategy.Level2HeuristicFunction);
fillerAction = WorldAction.NoAction;
isFirstTime = true;
strategyTimer = MaxStrategyTime;
calculatePathNextFrame = false;
}
// The center of the AI - get an action
override public List <WorldAction> GetAction(World world)
{
// The immediate action comes from level 1
WorldAction bestAction = WorldAction.NoAction;
// Update level 1 heuristic parameters
World.Player player = playerNum == 1 ? world.Player1 : world.Player2;
// Calculate new level 1 action if timer is up
if (decisionTimer <= 0)
{
ActionWithFiller decision = level1Searcher.ComputeBestAction(world, fillerAction, strategy.NextPathIndex);
bestAction = decision.Action;
fillerAction = decision.FillerAction;
decisionTimer = Level1StepSize;
// Otherwise do the filler action
}
else
{
bestAction = fillerAction;
// Check distance to path
bool doneWithPath = false;
if (strategy.SearchPath != null)
{
doneWithPath = strategy.NextPathIndex >= strategy.SearchPath.States.Count - 1;
}
// Calculate the path if this frame has been designated to it
if (calculatePathNextFrame)
{
// Run A*
Path path = level2Searcher.ComputeBestPath(blockWorld);
// Must be set before using the level 1 heuristic with a path
strategy.SearchPath = path;
strategy.NextPathIndex = 0;
calculatePathNextFrame = false;
// If no path is able to be calculated, then check again sooner than normal
if (path == null)
{
strategyTimer = NoPathFoundRefreshTimer;
}
}
else
{
// Compute a new strategy if the old one is no longer valid
SimplifiedWorld currentState = new SimplifiedWorld(world, playerNum);
if (isFirstTime ||
!previousState.IsEquivalent(currentState) ||
doneWithPath ||
dangerZoneShifted(world) ||
playerLeftPath(world, strategy.SearchPath) ||
strategyTimer <= 0 ||
world.IsTerminal())
{
if (isFirstTime)
{
previousState = currentState;
}
isFirstTime = false;
// Get reward and update QValues if learning
if (IsLearning)
{
float reward = SimplifiedWorld.Reward(previousState, strategy.Type, currentState);
QLearner.UpdateQValue(previousState, strategy.Type, currentState, reward);
// Don't learn once world is terminal
if (world.IsTerminal())
{
IsLearning = false;
}
}
// Get a new strategy
StrategyType newStrategy = QLearner.GetStrategy(currentState);
#if STRATEGY_PRINT
Debug.Log("Player " + playerNum.ToString() + " selects strategy: " + newStrategy.ToString());
#endif
strategy = Strategy.StrategyWithType(playerNum, newStrategy);
level1Searcher = new DiscreteAdversarialSearch(playerNum,
strategy.Level1Heuristic,
getFillerAction,
getNewPathIndex,
Level1StepSize,
4);
level2Searcher = new AStar(Level2MaxNodesInPrioQueue, Level2MaxExpansions, strategy.Level2CostFunction,
strategy.Level2GoalFunction, strategy.Level2HeuristicFunction);
// Create block world and danger zone
blockWorld = new BlockWorld(playerNum, world);
// Recalc danger zone
dangerZone = new DangerZone(opponentNum, world, blockWorld);
// Must be set before using the level 2 reward, cost, and goal functions
strategy.Level2DangerZone = dangerZone;
// Calculate the path in the next frame
calculatePathNextFrame = true;
// Speeds up framerate after player has died
if (world.IsTerminal())
{
calculatePathNextFrame = false;
}
// Reset previous state
previousState = currentState;
strategyTimer = MaxStrategyTime;
}
}
// Debug rendering of danger zone
#if DANGER_RENDER
dangerZone.Render(ResourceScript);
dangerZone.RenderPlayerBeliefs(ResourceScript);
#endif
}
// Advance path position
strategy.NextPathIndex = getNewPathIndex(player, strategy.NextPathIndex);
decisionTimer--;
strategyTimer--;
#if PATH_RENDER
if (strategy.SearchPath != null)
{
strategy.SearchPath.Render(ResourceScript, strategy.NextPathIndex);
}
#endif
// Return a single-valued list with the best action
return(new List <WorldAction>()
{
bestAction
});
}
