forked from RepoMirrors/baritone
much needed pathing overhaul
This commit is contained in:
parent
22bd5be5a9
commit
272dd79426
|
@ -25,7 +25,6 @@ import baritone.api.utils.BetterBlockPos;
|
|||
import baritone.pathing.calc.openset.BinaryHeapOpenSet;
|
||||
import baritone.pathing.movement.CalculationContext;
|
||||
import baritone.pathing.movement.Moves;
|
||||
import baritone.utils.Helper;
|
||||
import baritone.utils.pathing.BetterWorldBorder;
|
||||
import baritone.utils.pathing.Favoring;
|
||||
import baritone.utils.pathing.MutableMoveResult;
|
||||
|
@ -37,7 +36,7 @@ import java.util.Optional;
|
|||
*
|
||||
* @author leijurv
|
||||
*/
|
||||
public final class AStarPathFinder extends AbstractNodeCostSearch implements Helper {
|
||||
public final class AStarPathFinder extends AbstractNodeCostSearch {
|
||||
|
||||
private final Favoring favoring;
|
||||
private final CalculationContext calcContext;
|
||||
|
@ -55,14 +54,12 @@ public final class AStarPathFinder extends AbstractNodeCostSearch implements Hel
|
|||
startNode.combinedCost = startNode.estimatedCostToGoal;
|
||||
BinaryHeapOpenSet openSet = new BinaryHeapOpenSet();
|
||||
openSet.insert(startNode);
|
||||
bestSoFar = new PathNode[COEFFICIENTS.length];//keep track of the best node by the metric of (estimatedCostToGoal + cost / COEFFICIENTS[i])
|
||||
double[] bestHeuristicSoFar = new double[COEFFICIENTS.length];
|
||||
double[] bestHeuristicSoFar = new double[COEFFICIENTS.length];//keep track of the best node by the metric of (estimatedCostToGoal + cost / COEFFICIENTS[i])
|
||||
for (int i = 0; i < bestHeuristicSoFar.length; i++) {
|
||||
bestHeuristicSoFar[i] = startNode.estimatedCostToGoal;
|
||||
bestSoFar[i] = startNode;
|
||||
}
|
||||
MutableMoveResult res = new MutableMoveResult();
|
||||
Favoring favored = favoring;
|
||||
BetterWorldBorder worldBorder = new BetterWorldBorder(calcContext.world.getWorldBorder());
|
||||
long startTime = System.currentTimeMillis();
|
||||
boolean slowPath = Baritone.settings().slowPath.get();
|
||||
|
@ -75,10 +72,10 @@ public final class AStarPathFinder extends AbstractNodeCostSearch implements Hel
|
|||
int numNodes = 0;
|
||||
int numMovementsConsidered = 0;
|
||||
int numEmptyChunk = 0;
|
||||
boolean favoring = !favored.isEmpty();
|
||||
boolean isFavoring = !favoring.isEmpty();
|
||||
int timeCheckInterval = 1 << 6;
|
||||
int pathingMaxChunkBorderFetch = Baritone.settings().pathingMaxChunkBorderFetch.get(); // grab all settings beforehand so that changing settings during pathing doesn't cause a crash or unpredictable behavior
|
||||
boolean minimumImprovementRepropagation = Baritone.settings().minimumImprovementRepropagation.get();
|
||||
double minimumImprovement = Baritone.settings().minimumImprovementRepropagation.get() ? MIN_IMPROVEMENT : 0;
|
||||
while (!openSet.isEmpty() && numEmptyChunk < pathingMaxChunkBorderFetch && !cancelRequested) {
|
||||
if ((numNodes & (timeCheckInterval - 1)) == 0) { // only call this once every 64 nodes (about half a millisecond)
|
||||
long now = System.currentTimeMillis(); // since nanoTime is slow on windows (takes many microseconds)
|
||||
|
@ -136,21 +133,13 @@ public final class AStarPathFinder extends AbstractNodeCostSearch implements Hel
|
|||
throw new IllegalStateException(moves + " " + res.y + " " + (currentNode.y + moves.yOffset));
|
||||
}
|
||||
long hashCode = BetterBlockPos.longHash(res.x, res.y, res.z);
|
||||
if (favoring) {
|
||||
if (isFavoring) {
|
||||
// see issue #18
|
||||
actionCost *= favored.calculate(hashCode);
|
||||
actionCost *= favoring.calculate(hashCode);
|
||||
}
|
||||
PathNode neighbor = getNodeAtPosition(res.x, res.y, res.z, hashCode);
|
||||
double tentativeCost = currentNode.cost + actionCost;
|
||||
if (tentativeCost < neighbor.cost) {
|
||||
double improvementBy = neighbor.cost - tentativeCost;
|
||||
// there are floating point errors caused by random combinations of traverse and diagonal over a flat area
|
||||
// that means that sometimes there's a cost improvement of like 10 ^ -16
|
||||
// it's not worth the time to update the costs, decrease-key the heap, potentially repropagate, etc
|
||||
if (improvementBy < 0.01 && minimumImprovementRepropagation) {
|
||||
// who cares about a hundredth of a tick? that's half a millisecond for crying out loud!
|
||||
continue;
|
||||
}
|
||||
if (neighbor.cost - tentativeCost > minimumImprovement) {
|
||||
neighbor.previous = currentNode;
|
||||
neighbor.cost = tentativeCost;
|
||||
neighbor.combinedCost = tentativeCost + neighbor.estimatedCostToGoal;
|
||||
|
@ -159,12 +148,9 @@ public final class AStarPathFinder extends AbstractNodeCostSearch implements Hel
|
|||
} else {
|
||||
openSet.insert(neighbor);//dont double count, dont insert into open set if it's already there
|
||||
}
|
||||
for (int i = 0; i < bestSoFar.length; i++) {
|
||||
for (int i = 0; i < COEFFICIENTS.length; i++) {
|
||||
double heuristic = neighbor.estimatedCostToGoal + neighbor.cost / COEFFICIENTS[i];
|
||||
if (heuristic < bestHeuristicSoFar[i]) {
|
||||
if (bestHeuristicSoFar[i] - heuristic < 0.01 && minimumImprovementRepropagation) {
|
||||
continue;
|
||||
}
|
||||
if (bestHeuristicSoFar[i] - heuristic > minimumImprovement) {
|
||||
bestHeuristicSoFar[i] = heuristic;
|
||||
bestSoFar[i] = neighbor;
|
||||
if (getDistFromStartSq(neighbor) > MIN_DIST_PATH * MIN_DIST_PATH) {
|
||||
|
@ -182,28 +168,10 @@ public final class AStarPathFinder extends AbstractNodeCostSearch implements Hel
|
|||
System.out.println("Open set size: " + openSet.size());
|
||||
System.out.println("PathNode map size: " + mapSize());
|
||||
System.out.println((int) (numNodes * 1.0 / ((System.currentTimeMillis() - startTime) / 1000F)) + " nodes per second");
|
||||
double bestDist = 0;
|
||||
for (int i = 0; i < bestSoFar.length; i++) {
|
||||
if (bestSoFar[i] == null) {
|
||||
continue;
|
||||
Optional<IPath> result = bestSoFar(true, numNodes);
|
||||
if (result.isPresent()) {
|
||||
logDebug("Took " + (System.currentTimeMillis() - startTime) + "ms, " + numMovementsConsidered + " movements considered");
|
||||
}
|
||||
double dist = getDistFromStartSq(bestSoFar[i]);
|
||||
if (dist > bestDist) {
|
||||
bestDist = dist;
|
||||
}
|
||||
if (dist > MIN_DIST_PATH * MIN_DIST_PATH) { // square the comparison since distFromStartSq is squared
|
||||
logDebug("Took " + (System.currentTimeMillis() - startTime) + "ms, A* cost coefficient " + COEFFICIENTS[i] + ", " + numMovementsConsidered + " movements considered");
|
||||
if (COEFFICIENTS[i] >= 3) {
|
||||
System.out.println("Warning: cost coefficient is greater than three! Probably means that");
|
||||
System.out.println("the path I found is pretty terrible (like sneak-bridging for dozens of blocks)");
|
||||
System.out.println("But I'm going to do it anyway, because yolo");
|
||||
}
|
||||
System.out.println("Path goes for " + Math.sqrt(dist) + " blocks");
|
||||
return Optional.of(new Path(startNode, bestSoFar[i], numNodes, goal, calcContext));
|
||||
}
|
||||
}
|
||||
logDebug("Even with a cost coefficient of " + COEFFICIENTS[COEFFICIENTS.length - 1] + ", I couldn't get more than " + Math.sqrt(bestDist) + " blocks");
|
||||
logDebug("No path found =(");
|
||||
return Optional.empty();
|
||||
return result;
|
||||
}
|
||||
}
|
||||
|
|
|
@ -34,7 +34,7 @@ import java.util.Optional;
|
|||
*
|
||||
* @author leijurv
|
||||
*/
|
||||
public abstract class AbstractNodeCostSearch implements IPathFinder {
|
||||
public abstract class AbstractNodeCostSearch implements IPathFinder, Helper {
|
||||
|
||||
protected final int startX;
|
||||
protected final int startY;
|
||||
|
@ -53,7 +53,7 @@ public abstract class AbstractNodeCostSearch implements IPathFinder {
|
|||
|
||||
protected PathNode mostRecentConsidered;
|
||||
|
||||
protected PathNode[] bestSoFar;
|
||||
protected final PathNode[] bestSoFar = new PathNode[COEFFICIENTS.length];
|
||||
|
||||
private volatile boolean isFinished;
|
||||
|
||||
|
@ -63,13 +63,23 @@ public abstract class AbstractNodeCostSearch implements IPathFinder {
|
|||
* This is really complicated and hard to explain. I wrote a comment in the old version of MineBot but it was so
|
||||
* long it was easier as a Google Doc (because I could insert charts).
|
||||
*
|
||||
* @see <a href="https://docs.google.com/document/d/1WVHHXKXFdCR1Oz__KtK8sFqyvSwJN_H4lftkHFgmzlc/edit"></a>
|
||||
* @see <a href="https://docs.google.com/document/d/1WVHHXKXFdCR1Oz__KtK8sFqyvSwJN_H4lftkHFgmzlc/edit">here</a>
|
||||
*/
|
||||
protected static final double[] COEFFICIENTS = {1.5, 2, 2.5, 3, 4, 5, 10}; // big TODO tune
|
||||
protected static final double[] COEFFICIENTS = {1.5, 2, 2.5, 3, 4, 5, 10};
|
||||
|
||||
/**
|
||||
* If a path goes less than 5 blocks and doesn't make it to its goal, it's not worth considering.
|
||||
*/
|
||||
protected final static double MIN_DIST_PATH = 5;
|
||||
protected static final double MIN_DIST_PATH = 5;
|
||||
|
||||
/**
|
||||
* there are floating point errors caused by random combinations of traverse and diagonal over a flat area
|
||||
* that means that sometimes there's a cost improvement of like 10 ^ -16
|
||||
* it's not worth the time to update the costs, decrease-key the heap, potentially repropagate, etc
|
||||
* <p>
|
||||
* who cares about a hundredth of a tick? that's half a millisecond for crying out loud!
|
||||
*/
|
||||
protected static final double MIN_IMPROVEMENT = 0.01;
|
||||
|
||||
AbstractNodeCostSearch(int startX, int startY, int startZ, Goal goal, CalculationContext context) {
|
||||
this.startX = startX;
|
||||
|
@ -170,25 +180,43 @@ public abstract class AbstractNodeCostSearch implements IPathFinder {
|
|||
return Optional.ofNullable(mostRecentConsidered).map(node -> new Path(startNode, node, 0, goal, context));
|
||||
}
|
||||
|
||||
protected int mapSize() {
|
||||
return map.size();
|
||||
@Override
|
||||
public Optional<IPath> bestPathSoFar() {
|
||||
return bestSoFar(false, 0);
|
||||
}
|
||||
|
||||
@Override
|
||||
public Optional<IPath> bestPathSoFar() { // TODO cleanup code duplication between here and AStarPathFinder
|
||||
protected Optional<IPath> bestSoFar(boolean logInfo, int numNodes) {
|
||||
if (startNode == null || bestSoFar == null) {
|
||||
return Optional.empty();
|
||||
}
|
||||
for (int i = 0; i < bestSoFar.length; i++) {
|
||||
double bestDist = 0;
|
||||
for (int i = 0; i < COEFFICIENTS.length; i++) {
|
||||
if (bestSoFar[i] == null) {
|
||||
continue;
|
||||
}
|
||||
if (getDistFromStartSq(bestSoFar[i]) > MIN_DIST_PATH * MIN_DIST_PATH) { // square the comparison since distFromStartSq is squared
|
||||
return Optional.of(new Path(startNode, bestSoFar[i], 0, goal, context));
|
||||
double dist = getDistFromStartSq(bestSoFar[i]);
|
||||
if (dist > bestDist) {
|
||||
bestDist = dist;
|
||||
}
|
||||
if (dist > MIN_DIST_PATH * MIN_DIST_PATH) { // square the comparison since distFromStartSq is squared
|
||||
if (logInfo) {
|
||||
if (COEFFICIENTS[i] >= 3) {
|
||||
System.out.println("Warning: cost coefficient is greater than three! Probably means that");
|
||||
System.out.println("the path I found is pretty terrible (like sneak-bridging for dozens of blocks)");
|
||||
System.out.println("But I'm going to do it anyway, because yolo");
|
||||
}
|
||||
System.out.println("Path goes for " + Math.sqrt(dist) + " blocks");
|
||||
logDebug("A* cost coefficient " + COEFFICIENTS[i]);
|
||||
}
|
||||
return Optional.of(new Path(startNode, bestSoFar[i], numNodes, goal, context));
|
||||
}
|
||||
}
|
||||
// instead of returning bestSoFar[0], be less misleading
|
||||
// if it actually won't find any path, don't make them think it will by rendering a dark blue that will never actually happen
|
||||
if (logInfo) {
|
||||
logDebug("Even with a cost coefficient of " + COEFFICIENTS[COEFFICIENTS.length - 1] + ", I couldn't get more than " + Math.sqrt(bestDist) + " blocks");
|
||||
logDebug("No path found =(");
|
||||
}
|
||||
return Optional.empty();
|
||||
}
|
||||
|
||||
|
@ -205,4 +233,8 @@ public abstract class AbstractNodeCostSearch implements IPathFinder {
|
|||
public BetterBlockPos getStart() {
|
||||
return new BetterBlockPos(startX, startY, startZ);
|
||||
}
|
||||
|
||||
protected int mapSize() {
|
||||
return map.size();
|
||||
}
|
||||
}
|
||||
|
|
Loading…
Reference in New Issue