void POCMAN::GeneratePreferred(const STATE& state, const HISTORY& history,
vector<int>& actions, const STATUS& status) const
{
const POCMAN_STATE& pocstate = safe_cast<const POCMAN_STATE&>(state);
if (history.Size())
{
int action = history.Back().Action;
int observation = history.Back().Observation;
// If power pill and can see a ghost then chase it
if (pocstate.PowerSteps > 0 && (observation & 15 != 0))
{
for (int a = 0; a < 4; ++a)
if (CheckFlag(observation, a))
actions.push_back(a);
}
// Otherwise avoid observed ghosts and avoid changing directions
else
{
for (int a = 0; a < 4; ++a)
{
COORD newpos = NextPos(pocstate.PocmanPos, a);
if (newpos.Valid() && !CheckFlag(observation, a)
&& COORD::Opposite(a) != action)
actions.push_back(a);
}
}
}
}
bool ROCKSAMPLE::LocalMove(STATE& state, const HISTORY& history,
int stepObs, const STATUS& status) const
{
_unused(status);
ROCKSAMPLE_STATE& rockstate = safe_cast<ROCKSAMPLE_STATE&>(state);
int rock = Random(NumRocks);
rockstate.Rocks[rock].Valuable = !rockstate.Rocks[rock].Valuable;
if (history.Back().Action > E_SAMPLE) // check rock
{
rock = history.Back().Action - E_SAMPLE - 1;
int realObs = history.Back().Observation;
// Condition new state on real observation
int newObs = GetObservation(rockstate, rock);
if (newObs != realObs)
return false;
// Update counts to be consistent with real observation
if (realObs == E_GOOD && stepObs == E_BAD)
rockstate.Rocks[rock].Count += 2;
if (realObs == E_BAD && stepObs == E_GOOD)
rockstate.Rocks[rock].Count -= 2;
}
return true;
}
void QNODE::DisplayValue(HISTORY& history, int maxDepth, ostream& ostr, const double *qvalue) const
{
history.Display(ostr);
if (qvalue) {
ostr << "q=" << *qvalue;
}
ImmediateReward.Print(": r=", ostr);
Observation.Print(", o=", ostr);
ostr << std::endl;
for (int observation = 0; observation < NumChildren; observation++)
{
if (Children[observation])
{
std::stringstream ss;
ss << "\t\t\t#" << observation;
// Children[observation]->GetCumulativeReward().Print(ss.str().c_str(), ostr);
}
}
if (history.Size() >= maxDepth)
return;
for (int observation = 0; observation < NumChildren; observation++)
{
if (Children[observation])
{
history.Back().Observation = observation;
Children[observation]->DisplayValue(history, maxDepth, ostr);
}
}
}
bool ROOMS::LocalMove(STATE &state, const HISTORY &history, int) const
{
ROOMS_STATE rstate = safe_cast<ROOMS_STATE &>(state);
if (GetObservation(rstate) == history.Back().Observation) {
return true;
}
return false;
}
void QNODE::DisplayPolicy(HISTORY& history, int maxDepth, ostream& ostr) const
{
history.Display(ostr);
ostr << ": " << Value.GetValue() << " (" << Value.GetCount() << ")\n";
if (history.Size() >= (uint) maxDepth)
return;
for (int observation = 0; observation < NumChildren; observation++)
{
if (Children[observation])
{
history.Back().Observation = observation;
Children[observation]->DisplayPolicy(history, maxDepth, ostr);
}
}
}
bool ROBOT_NAVIGATION::LocalMove(STATE& state, const HISTORY& history, int stepObs, const STATUS& status) const
{
ROBOT_STATE& robotstate = safe_cast<ROBOT_STATE&>(state);
/*
* We assume perfect observation of the grid cells immediately surrounding the agent.
* As a result, the robot can observe the wall configurations in surrounding squares,
* but not its own actual location.
*/
int obs = 0;
for (int i=0; i<8; i++)
{
int obsx = robotstate.X + ROBOT_NAVIGATION::DeltaObs[i][0];
int obsy = robotstate.Y + ROBOT_NAVIGATION::DeltaObs[i][1];
if (ROBOT_NAVIGATION::Map[obsx][obsy] == 1)
obs += 1 << i;
}
return (obs == history.Back().Observation);
}
void QNODE::DisplayPolicy(HISTORY& history, int maxDepth, ostream& ostr) const
{
history.Display(ostr);
ImmediateReward.Print("r=", ostr);
Observation.Print(", o=", ostr);
ostr << std::endl;
if (history.Size() >= maxDepth)
return;
for (int observation = 0; observation < NumChildren; observation++)
{
if (Children[observation])
{
history.Back().Observation = observation;
Children[observation]->DisplayPolicy(history, maxDepth, ostr);
}
}
}
bool POCMAN::LocalMove(STATE& state, const HISTORY& history,
int stepObs, const STATUS& status) const
{
_unused(stepObs);
_unused(status);
POCMAN_STATE& pocstate = safe_cast<POCMAN_STATE&>(state);
int numGhosts = Random(1, 3); // Change 1 or 2 ghosts at a time
for (int i = 0; i < numGhosts; ++i)
{
int g = Random(NumGhosts);
pocstate.GhostPos[g] = COORD(
Random(Maze.GetXSize()),
Random(Maze.GetYSize()));
if (!Passable(pocstate.GhostPos[g])
|| pocstate.GhostPos[g] == pocstate.PocmanPos)
return false;
}
COORD smellPos;
for (smellPos.X = -SmellRange; smellPos.X <= SmellRange; smellPos.X++)
{
for (smellPos.Y = -SmellRange; smellPos.Y <= SmellRange; smellPos.Y++)
{
COORD pos = pocstate.PocmanPos + smellPos;
if (smellPos != COORD(0, 0) &&
Maze.Inside(pos) &&
CheckFlag(Maze(pos), E_SEED))
pocstate.Food[Maze.Index(pos)] = Bernoulli(FoodProb * 0.5);
}
}
// Just check the last time-step, don't check for full consistency
if (history.Size() == 0)
return true;
int observation = MakeObservations(pocstate);
return history.Back().Observation == observation;
}