public QState GetState()
{
var winner = _ball.IsTerminal();
float reward;
bool terminal;
//var terminal = winner.HasValue;
//var reward = terminal ? (winner.Value == Side ? 1 : 0) : 0;
var b = PongGame.BoundsFromTransform(_ball.transform);
var controller = PongGame.BoundsFromTransform(transform);
controller.size += new Vector3(0.2f, 0, 0);
PongGame.DebugDrawBounds(controller, Color.blue);
if (b.Intersects(controller))
{
reward = 1;
terminal = true;
if (PongBenchmark.Running)
{
terminal = winner.HasValue;
}
}
else
{
terminal = winner.HasValue;
reward = terminal ? (winner.Value == Side ? 1 : 0) : 0;
}
var bp = _ball.transform.position;
var rbp = bp - transform.position;
var gbp = _grid.Locate(bp);
var bpy = gbp.HasValue ? gbp.Value.y : -1;
_grid.Populate((bo, c) => {
//var x = bo.center.x;
//var v = bo.Contains(new Vector3(x, _game.Border.min.y)) || bo.Contains(new Vector3(x, _game.Border.max.y)) ? 0.3f : 0; // walls
var ham = gbp.HasValue ? HammingDistance(gbp.Value, c) : int.MaxValue; // Hamming distance
var v = ham <= 0 ? 1f : ham <= 1 ? 0.5f : 0; // ball
//v = bo.Contains(bp + _ball.Velocity.normalized * 2) ? 150f : v;
//v = bo.Intersects(controller) ? 100 : v;
return(v);
});
var state = _grid.Matrix;
return(new QState(new[] { state }, _vect.Clone(), reward, terminal));
}
public QState GetState()
{
_grid.SetAll(1f);
for (int i = -2; i < 20; i++)
{
var point = Track.GetPointAtDistance(_distanceTravelled + i * 0.8f);
var coordinates = _grid.Locate(point);
if (coordinates.HasValue)
{
_grid[coordinates.Value] = 0f;
}
}
var linear = Vector <float> .Build.Dense(
_velocityBin.Get(Velocity / 20f)
.Concat(_forceBin.Get(Mathf.Abs(Force))).ToArray()
);
// _vector[0] = Velocity / 20f;
// _vector[1] = Mathf.Abs(Force);
// var reward = Mathf.Abs(DistanceTravelled - StartPosition) - lastReward > 0.001f ? 0.5f : 0f;
var reward = Velocity / 20f;
reward += Mathf.Abs(Force) * -0.2f;
// reward = !OnTrack && Mathf.Abs(Force) > 1f ? 0f : reward;
// reward += DistanceTravelled - StartPosition > 80 ? 80 / LapTime : 0;
if (_standStillTicks > 20)
{
reward = 0;
}
var terminal =
_distanceTravelled - StartPosition > Track.length ||
!_onTrack ||
_standStillTicks > 200;
if (QAIManager.CurrentMode == QAIMode.Testing)
{
terminal = _distanceTravelled - StartPosition > Track.length ||
_standStillTicks > 200;
}
var state = new QState(
new [] { _grid.Matrix },
linear.Clone(),
// !terminal ? 0 : (DistanceTravelled - StartPosition) / (Track.length/2),
reward,
terminal);
if (Velocity < 1f)
{
_standStillTicks++;
}
else
{
_standStillTicks = 0;
}
return(state);
}