Program a Racing Car

36

11

CONGRATULATIONS to @kuroineko. Wins the bounty for excellent speed (672 moves) on the Gauntlet track.

LEADER:* Nimi scoring a lightweight 2129. Other entries are larger but showing some serious speed.

*Leader may change due to later entries.

Your task is to write a small program that can drive a racing car fast.

Rules

Your program will read in an image of the track. You can start your car on any yellow pixel, and you must finish by crossing any black pixel. The path of your car must be only on the grey ((c,c,c) where 30 <= c <= 220) track.

Your car will move in a straight line each turn with velocity v made up of integers vx and vy (starting with (0,0)). At the start of each turn your program can change vx and vy such that:

abs(vx2-vx1) + abs(vy2-vy1) <= 15

Update: Increased maximum acceleration to 15.

Your program will then plot a straight line from your current location to (location + v) in white with a blue dot at the start. If a pixel under this line is black, you have finished the race. Otherwise, if all the pixels under that line are grey or yellow, you may continue to the next turn.

Your program must output the track image with your path in white and blue added.

Additional Output (added 2015-01-15):

If you wish to compete for the win or the bonus, your program should also output your list of points (the blue dots) for the City or Gauntlet respectively. Include the list of points with your answer (for verification). The points should look like: (x0,y0), (x1,y1), ... (xn,yn). You can use white space freely including '\n' characters to fit the data on the page.

You may use third party image reading and writing, line drawing and pixel access libraries. You can't use path-finding libraries. If you wish you can convert the PNG images into other graphics formats (such as GIF, JPG, BMP) if you need to.

A few tracks to drive on

A simple track to start out:

Simple Track

A Race Track:

Race Track

An Obstacle Course:

Obstacle Course

The City:

The City

Nightmare Track: The Gauntlet (if the others are too easy)

The Gauntlet

Scoring

Your score will be based on your result on the City track. Points are equal to your program length in bytes plus 10 points for each turn your racing car took to finish. Lowest score wins. Please include your City track run image with your answer - we would like to see your driving style.

Please use a title for your answer in the format:

<Racing Driver or Team Name> <Language> <Score> eg: Slowpoke Perl 5329

Your program must be able to drive on any track image following the above rules. You must not hardcode the optimum path or any parameters of the test tracks. The other standard loopholes apply.

Similar challenges

This is a similar puzzle to one posed by Martin: To Vectory! – The Vector Racing Grand Prix. This puzzle has a number of differences:

  • Driving through walls is not allowed.
  • You can use unlimited memory and time.
  • You do not have to run somebody else's code on your computer.
  • You do not have to download anything except one image.
  • The size of your code counts in scoring. Smaller is better.
  • You plot your solution on the track image.
  • You can easily create your own tracks with a paint package.
  • The higher resolution encourages more realistic braking and cornering.
  • Acceleration of 15 creates about 450 possibilities per turn. This makes BFS less feasible and encourages new interesting algorithms.

This puzzle should inspire a new round of programmers to try solutions and allow programmers with old solutions to rethink them in the new environment.

Logic Knight

Posted 2014-12-06T07:27:26.907

Reputation: 6 622

Just an idea: I fear a Gauntlet track solution, given just as a picture file is hard for you to test against your acceleration constraint. I think verifying any NEW-BOUNTY-solution would be easier for you, if you insist on a listing of the path polygon points like so: (x0, y0) (x1, y1) ... (xn, yn). Where n is the turn count found. – Bob Genom – 2015-01-14T20:37:58.457

@Bob. Good idea. It is too hard to judge from the code or the resulting image by eye. I will update the rules. It should not be to hard to update programs to output these points. I will write a small script to verify the points too. – Logic Knight – 2015-01-15T06:37:24.620

I tried a variant based on RRTs. Noncompeting - didn't even try to golf it. Works... interestingly? Works very well on many-branching courses, but doesn't work well at all on anything with a lot of doubling back on itself. (It has to wait until it randomly chooses a point nearer the current leaves than the previous part of the course, not including velocity, before it continues.). On the gauntlet it never even gets out of the starting area :( – TLW – 2016-09-01T22:44:01.690

@xnor duplicate enough but not perfect, I have to add: This question uses graphical input (leading to much larger boards) and allows more acceleration. The BFS implementation would probably time out here. There are also no opponents. – John Dvorak – 2014-12-06T08:02:32.100

This challenge also disallows tunneling through walls. Which brings me to my question: What line drawing algorithm do we have to use? If we have leeway here, how much? I feel like this might be abusible somewhat - especilally if "hand-drawn" "lines" are allowed.

– John Dvorak – 2014-12-06T08:06:14.587

I like Martin's challenge but I was aiming for something different here. There is a code-golf aspect which rewards small clever algorithms. There is also graphical output which stimulates competition. I hope this question will be allowed - I am looking forward to some clever answers. – Logic Knight – 2014-12-06T08:07:54.453

@Jan A straight line is required with no anti-aliasing. Built-in line drawing functions are allowed, or you can plot a line using Bresenham's line algorithm. – Logic Knight – 2014-12-06T08:10:52.557

2If the duplicate thing turns out to not be a problem, I think you should add a lot more tracks to judge with. There's many degrees of hard-coding. This makes it hard to draw the line between a program that optimizes for a map with general featured line "roads" and one heavily tailored to this specific map. – xnor – 2014-12-06T08:36:26.550

@MartinBüttner the grey is well-specified. As for yellow, I'd assume #FF0 – John Dvorak – 2014-12-06T10:48:04.640

Good idea. Perhaps make the score equal to code size plus the average of several tracks times 10. I am a lousy pixel artist. Anybody like to come up with an interesting track? – Logic Knight – 2014-12-06T10:55:55.630

@Martin Yes the yellow, black, and white have the usual unique RGB values. Grey is is much more broad and specified in the question. – Logic Knight – 2014-12-06T11:06:29.687

"You must not hard code the optimum path or any of the parameters of the test tracks" Can we tell the program the coordinates of the finish? Otherwise it's a maze exploring problem, which makes the title rather misleading. – Level River St – 2014-12-06T11:12:08.790

@Steve Your program must find a finish location with internal logic after it has read in the track image file, then (I assume) plan a path from a start to the finish. I guess it is a cross between rally driving and orienteering. The path-finding aspect is certainly important, but so is the optimal braking and cornering. – Logic Knight – 2014-12-06T11:18:13.400

I think the global idea is fun, but the golf scoring discourages sophistication. Basically your test track will cost 3000 points to a bot moving at constant speed, and beating that will only reduce the cost to maybe 1500, which can probably not be achieved with 1500 or less extra characters in the source. – None – 2014-12-08T15:14:52.697

@kuroineko, This is a central part of this puzzle. The best scores will be from small clever algorithms. Complex programs (although with a perfect path and car control) will attract a size penalty. You might try (as Bob has done below) a simple but effective program, then improve it and shrink it to be competitive. Of course you could also build a larger clever program and display your great car control, which may attract upvotes. Good results will always get a vote from me. – Logic Knight – 2014-12-08T16:13:33.507

well the challenge is interresting anyway, but still the golf part practically rules out languages like C++ that cannot read png images natively. – None – 2014-12-09T13:38:56.037

1I should have made this obvious in the question, but you can use any third party libraries to read and write images, plot lines and dots etc. You may not use a path-finding library though. Like all golf-like questions, verbose languages (like Java) will suffer, but you can take satisfaction in being the best in your language group. – Logic Knight – 2014-12-09T14:13:51.077

Answers

6

TS#1 - Haskell - 1699 + 430 = 2129

Tutu sibling #1

Very much the same as the original Tutu racer, except it uses a thickness of 3 for the bloated path and the 2nd A* (speed-pos space) goes with a constant heuristic of 1. The input picture is not passed as a command line argument anymore, it must be named i. The name of the output picture is o. The program prints the calculated points on the path as a list of x,y pairs (original output is a single line):

[(6,7),(20,6),(49,5),(92,5),(124,4),(141,3),(148,7),(155,26),(172,49),
(189,70),(191,91),(179,111),(174,124),(184,137),(209,150),(244,168),
(279,171),(302,176),(325,196),(350,221),(367,239),(369,257),(360,272),
(363,284),(381,296),(408,314),(433,329),(458,329),(480,318),(492,312),
(504,321),(519,341),(526,364),(523,392),(514,416),(507,427),(511,435),
(529,442),(558,445),(581,456),(592,470),(592,488),(592,513),(606,537)] 

I could save a lot of bytes when I start to remove all the map and set data structures and replace them with simple linked lists. Just the two import statements would save 60 bytes. However, it would slow down the program so that waiting for a result is pure pain. This versions takes over 45min for The City, compared to the 7s of the original. I’ll stop here trading bytes for executing speed.

import Codec.Picture
import Codec.Picture.RGBA8
import Codec.Picture.Canvas
import qualified Data.Map as M
import qualified Data.Set as S
import qualified Data.PSQueue as Q
m(a,b)(c,d)|a==c||b==d=2|t=3;n=Nothing;q=PixelRGBA8;s=abs;t=1<2;u=signum;z=255;fl=S.fromList;(#)=M.insert
main=do
 i<-readImageRGBA8"i";let(Right c)=imageToCanvas i;j=canvasWidth c-1;gY=canvasHeight c-1;v(x,y)|all(==0)[r,g,b]=3|r+g==510&&b==0=2|r==g&&r==b&&29<r&&r<221=0|t=1 where(PixelRGBA8 r g b _)=getColor x y c
 let s':_=[(x,y)|x<-[0..j],y<-[0..gY],v(x,y)==2];n8 p@(x,y)=filter((/=1).v)$if y*x==0||y==gY||x==j then[p]else[(a,b)|a<-[x-1..x+1],b<-[y-1..y+1],a/=x||b/=y];r=s':aS(fl.n8)m((==3).v)s';f=concatMap n8;p=head r;w=map fst$(p,(0,0)):aS(\((a,b),(h,i))->fl[(e,(h+j,i+k))|j<-[-15..15],k<-[s j-15..15-s j],not$all(==0)[j,k,h,i],let e=(a+h+j,b+i+k),S.member e(fl$f$f$f r),all((/=1).v)(br(a,b)e)])(\_ _->99)((==last r).fst)(p,(0,0))
 writePng"o"$canvasToImage$foldl(\e((a,b),(c,d))->setColor a b(q 0 0 z z)$drawLine a b c d(q z z z z)e)c(zip w(tail w));print w
br q@(i,j)r@(k,l)=w q$f`div`2where w p@(y,x)e|p==r=[p]|e-o<0=p:w(y+g,x+h)(e-o+f)|t=p:w(y+m,x+n)(e-o);a=s$l-j;b=s$k-i;h=u$l-j;g=u$k-i;(n,m,o,f)|a>b=(h,0,b,a)|t=(0,g,a,b)
data A a c=A{a::S.Set a,h::Q.PSQ a c,k::M.Map a c,p::M.Map a a,w::Maybe a}
aS g d o u=b$w s where b(Just j)=(reverse$takeWhile(/=u)$iterate(p s M.!)j);s=l$A S.empty(Q.singleton u 0)(M.singleton u 0)M.empty n;i x y v s=s{p=y#x$p s,k=y#v$k s,h=Q.insert y(v+1)$h s};l s=b$Q.minView$h s where b(Just(x Q.:->_,w'))|o x=s{w=Just x}|t=l$foldl(r x)(s{h=w',a=S.insert x(a s)})$S.toList$g x S.\\a s;b _=s;r x s y=b$Q.lookup y$h s where v=k s M.!x+d x y;b Nothing=i x y v s;b _|v<k s M.!y=i x y v s|t=s

The code needs the -XNoMonomorphismRestriction flag to compile.

The City - TS#1 - 43 steps TS#1 - 43 steps

nimi

Posted 2014-12-06T07:27:26.907

Reputation: 34 639

Acc limit checked. Avg acceleration = 13.627 too showing close to maximum acceleration, cornering, and braking. – Logic Knight – 2015-01-19T14:24:35.953

Yet another incentive for me to switch to C++. Brute force will prevail one day. I know it will! – None – 2015-01-30T10:47:31.437

12

FirstRacer Java (5825 + 305*10 = 8875)

Just to have a start. Needs 305 segments on City.

This Java program does it pipelined:

  1. read the image
  2. A* (A star)
  • 2.1 Build the A* search landscape.
  • 2.2. Track back looking only for the best *direct* 8 neighbor cells (N,S,E,W,NE,NW,SE,SW). This finds the shortest track t0 pixelwise.
  1. stay on t0 and optimize it for speed by eliminating pixels. Fullfilling the constraint by never be faster than 7 (this translates to keep at least every 7th pixel).
  2. draw the track into the image
  3. show the resulting image.
package race;

import java.awt.Color;
import java.awt.Graphics;
import java.awt.Graphics2D;
import java.awt.Point;
import java.awt.image.BufferedImage;
import java.io.File;
import java.io.IOException;
import java.util.Vector;

import javax.imageio.ImageIO;
import javax.swing.JFrame;
import javax.swing.WindowConstants;

public class AStar {

    private static BufferedImage img;
    private static int Width;
    private static int Height;
    private static int[][] cost;
    private static int best=Integer.MAX_VALUE;
    private static Point pBest;

    public static void main(String[] args) throws IOException {
        String file = "Q46YG.png";
        img = read(file);
        Width=img.getWidth();
        Height=img.getHeight();

        Vector<Point> track = astar();
        track = optimize(track);
        draw(track);
        System.out.println(10 * track.size());

        JFrame frame = new JFrame(file) {
            public void paint(Graphics g) {
                setSize(Width+17, Height+30+10);
                g.drawImage(img,8,30,null);
            }
        };
        frame.setDefaultCloseOperation(WindowConstants.EXIT_ON_CLOSE);
        frame.setVisible(true);
    }

    private static Vector<Point> optimize(Vector<Point> track) {
        Vector<Point> opt=new Vector<Point>();
        Point p0 = track.get(0);
        Point p1 = track.get(1);
        int v=0;
        opt.add(p0);
        int vx0=p1.x-p0.x, vy0=p1.y-p0.y;
        for (int i = 2; i < track.size(); i++) {
            Point p = track.get(i);
            if (v<7 && vx0==p.x-p1.x && vy0==p.y-p1.y) {
                v++;
            } else {
                v=0;
                opt.add(p1);
                vx0=p.x-p1.x;
                vy0=p.y-p1.y;
            }
            p1=p;
        }
        opt.add(p1);
        return opt;
    }

    private static void draw(Vector<Point> track) {
        Graphics2D g = img.createGraphics();
        Point p0 = track.get(0);
        for (int i = 1; i < track.size(); i++) {
            Point p1 = track.get(i);
            g.setColor(Color.WHITE);
            g.drawLine(p0.x, p0.y, p1.x, p1.y);
            img.setRGB(p0.x, p0.y, 0xff0000ff);
            img.setRGB(p1.x, p1.y, 0xff0000ff);
            p0=p1;
        }
    }

    private static Vector<Point> astar() {
        Vector<Point> v0=findStart();
        for(int i=0; ; i++) {
            Vector<Point> v1=next(v0);
            if (v1.size()==0) break;
            v0=v1;
        }
        Vector<Point> track=new Vector<Point>();
        Point p0 = pBest;
        int x0=p0.x, y0=p0.y;
        int c0=cost[x0][y0];
        while(true) {
            int x=x0, y=y0;
            track.add(0, new Point(x, y));
            for (int x1 = x-1; x1 <= x+1; x1++) {
                for (int y1 = y-1; y1 <= y+1; y1++) {
                    int i1=getInfo(x1, y1);
                    if ((i1&2)==2) {
                        int c=cost[x1][y1];
                        if (c0>c) {
                            c0=c;
                            x0=x1;
                            y0=y1;
                        }
                    }
                }
            }
            if(x0==x &&y0==y) break;
        }
        return track;
    }

    private static Vector<Point> next(Vector<Point> v0) {
        Vector<Point> v1=new Vector<Point>();
        for (Point p0 : v0) {
            int x=p0.x, y=p0.y;
            int c0=cost[x][y];
            for (int x1 = x-1; x1 <= x+1; x1++) {
                for (int y1 = y-1; y1 <= y+1; y1++) {
                    int i1=getInfo(x1, y1);
                    if ((i1&2)==2) {
                        int c1=c0+1414;
                        if (x1==x || y1==y) {
                            c1=c0+1000;
                        }
                        int c=cost[x1][y1];
                        if (c1<c) {
                            cost[x1][y1]=c1;
                            Point p1=new Point(x1, y1);
                            v1.add(p1);
                            if (i1==3) {
                                if (best>c1) {
                                    best=c1;
                                    pBest=p1;
                                }
                            }
                        }
                    }
                }
            }

        }
        return v1;
    }

    private static Vector<Point> findStart() {
        cost=new int[Width][Height];
        Vector<Point> v=new Vector<Point>();
        for (int i = 0; i < Width; i++) {
            for (int j = 0; j < Height; j++) {
                if (getInfo(i,j)==1) {
                    cost[i][j]=0;
                    v.add(new Point(i, j));
                } else {
                    cost[i][j]=Integer.MAX_VALUE;
                    pBest=new Point(i, j);
                }
            }
        }
        return v;
    }

    /**
     * 1: You can start your car on any yellow pixel, 
     * 3: and you must finish by crossing any black pixel. 
     * 2: The path of your car must be only on the grey ((c,c,c) where 30 <= c <= 220) track.
     * 0: else
     * 
     * @param x
     * @param y
     * @return
     */
    private static int getInfo(int x, int y) {
        if (x<0 || x>=Width || y<0 || y>=Height) return 0;
        int rgb = img.getRGB(x, y);
        int c=0;
        switch (rgb) {
        case 0xffffff00: c=1; break;
        case 0xff000000: c=3; break;
        default: 
            int r=0xff&(rgb>>16);
            int g=0xff&(rgb>> 8);
            int b=0xff&(rgb>> 0);
            if (30<=r&&r<=220&&r==g&&g==b) c=2;
        }
        return c;
    }

    private static BufferedImage read(String file) throws IOException {
        File img = new File("./resources/"+file);
        BufferedImage in = ImageIO.read(img);
        return in;
    }

}

The City

I think the Race Track gives you a better impression how FirstRacer works. Race Track

Bob Genom

Posted 2014-12-06T07:27:26.907

Reputation: 846

That thing that your car does in the city in the middle of the lower block... doesn't seem optimal. – John Dvorak – 2014-12-08T01:52:39.923

3Despite the simple algorithm, your car does seem to take a good line through the corners. Now if you can just get out of first gear... – Logic Knight – 2014-12-08T02:12:57.320

I calculate your initial score to be 5825 + 305*10 = 8875. – Logic Knight – 2014-12-08T02:28:42.457

@JanDvorak Yes,this version is a kind of dump. – Bob Genom – 2014-12-08T14:47:08.753

@CarpetPython The next version (if challenged) will use the gear to have only 276 (or less) turns. Thank you for calculating my score. – Bob Genom – 2014-12-08T14:52:20.560

11

pighead PHP (5950+470 = 6420)

Yet another (pigheaded) BFS variation, with some preprocessing to trim down down search space.

<?php
define ("ACCEL_MAX", 15);
define ("TILE_SIZE_MAX", 2*floor (ACCEL_MAX/2)-1);
define ("TILE_SIZE_MIN", 1);

class Point {
    function __construct ($x=0, $y=0)
    {
        $this->x = (float)$x;
        $this->y = (float)$y;
    }

    function add ($v)
    {
        return new Point ($this->x + $v->x, $this->y + $v->y);
    }
}

class Tile {
    public $center;
    private static $id = 0;

    public function __construct ($corner_x, $corner_y, $size, $type)
    {
        $this->type = $type;
        $this->id = ++self::$id;
        $half = round ($size/2);
        $this->center = new Point ($corner_x+$half, $corner_y+$half));
        for ($x = 0 ; $x != $size ; $x++)
        for ($y = 0 ; $y != $size ; $y++)
            Map::$track[$x+$corner_x][$y+$corner_y] = 0;
        Map::$tile_lookup[$this->center->x][$this->center->y] = $this;
    }

    public function can_reach ($target)
    {
        if (isset($this->reachable[$target->id])) return $this->reachable[$target->id];
        $ex = $target->center->x;
        $ey = $target->center->y;
        $ox = $this->center->x;
        $oy = $this->center->y;
        $sx = $ex - $ox;
        $sy = $ey - $oy;
        $range = max (abs ($sx), abs ($sy));
        if ($range == 0) return false;
        $reachable = true;
        for ($s = 1 ; $s != $range ; $s++)
            if (!isset (Map::$track[$ox + $s/$range*$sx][$oy + $s/$range*$sy]))
            {
                $reachable = false;
                break;
            }
        return $this->reachable[$target->id] = $target->reachable[$this->id] = $reachable;
    }
}

class Node {
    public $posx  , $posy  ;
    public $speedx, $speedy;
    private $parent;

    public function __construct ($posx, $posy, $speedx, $speedy, $parent)
    {
        $this->posx = $posx;
        $this->posy = $posy;
        $this->speedx = $speedx;
        $this->speedy = $speedy;
        $this->parent = $parent;
    }

    public function path ()
    {
        $res = array();
        for ($node = $this ; $node != null ; $node = $node->parent)
        {
            array_unshift ($res, new Point ($node->posx, $node->posy));
        }
        return $res;
    }
}

class Map {
    public static $track;       // map of track pixels
    public static $tile_lookup; // retrieve tile from a position

    private static $tiles;        // all track tiles
    private static $sx, $sy;      // map dimensions
    private static $cell;         // cells of the map
    private static $start;        // starting point
    private static $acceleration; // possible acceleration values
    private static $img; // output image
    private static $output_name;

    const GOAL  = 0;  // terrain types
    const START = 1;
    const TRACK = 2;

    private static function create_tile ($cx, $cy, $size)
    {
        for ($x = $cx ; $x != $cx + $size ; $x++)
        for ($y = $cy ; $y != $cy + $size ; $y++)
            if (!isset (self::$track[$x][$y]) || !self::$track[$x][$y]) return false;
        for ($x = $cx ; $x != $cx + $size ; $x++)
        for ($y = $cy ; $y != $cy + $size ; $y++)
            self::$track[$x][$y] = 0;
//Trace::msg ("track tile $cx $cy $size");
        return new Tile ($cx, $cy, $size, self::TRACK);
    }

    public static function init ($filename)
    {
        // read map definition
        $img = imagecreatefrompng ($filename) or die ("could not read $filename");
        self::$img = $img;
        self::$output_name = "_".$filename;
        self::$sx = imagesx ($img);
        self::$sy = imagesy ($img);

        for ($x = 0 ; $x != self::$sx ; $x++)
        for ($y = 0 ; $y != self::$sy ; $y++)
        {
            $color = imagecolorat ($img, $x, $y) & 0xFFFFFF;
            if      ($color  ==        0) self::$tiles[]                  = new Tile ($x, $y, 1, Map::GOAL);
            else if ($color  == 0xFFFF00) self::$tiles[] = self::$start[] = new Tile ($x, $y, 1, Map::START);
            else
            {
                $r = ($color >> 16) & 0xFF;
                $g = ($color >>  8) & 0xFF;
                $b =  $color        & 0xFF;
                if ($r == $g && $r == $b && $r >= 30 && $r <= 220) @self::$track[$x][$y] = 1;
            }
        }

        for ($size = TILE_SIZE_MAX ; $size >= TILE_SIZE_MIN ; $size--)
        for ($x = 0 ; $x != self::$sx ; $x++)
        for ($y = 0 ; $y != self::$sy ; $y++)
        {
            $tile = self::create_tile ($x, $y, $size);
            if ($tile) self::$tiles[] = $tile;
        }

        self::$acceleration = array();
        for ($x = -ACCEL_MAX ; $x <= ACCEL_MAX ; $x++)
        for ($y = -ACCEL_MAX ; $y <= ACCEL_MAX ; $y++)
        {
            if (abs ($x) + abs ($y) <= ACCEL_MAX) self::$acceleration[] = new Point ($x, $y);
        }
    }

    public static function solve ()
    {
        $res = $border = $present = array();
        foreach (self::$start as $start)
        {
            $border[] = new Node ($start->center->x, $start->center->y, 0, 0, null);
            $present[$start->center->x." ".$start->center->y." 0 0"] = 1;
        }
        while (count ($border))
        {
            $node = array_shift ($border);
            $px = $node->posx;
            $py = $node->posy;
            $vx = $node->speedx;
            $vy = $node->speedy;
            $current = self::$tile_lookup[$px][$py];
            foreach (self::$acceleration as $a)
            {
                $nvx = $vx + $a->x;
                $nvy = $vy + $a->y;
                $npx = $px + $nvx;
                $npy = $py + $nvy;
                @$tile = self::$tile_lookup[$npx][$npy];
                if (!$tile || !$tile->can_reach ($current)) continue;
                if ($tile->type == self::GOAL)
                {
                    $end = new Node ($npx, $npy, $nvx, $nvy, $node);
                    $res = $end->path ();
                    $ox = $res[0]->x;
                    $oy = $res[0]->y;
                    for ($i = 1 ; $i != count ($res) ; $i++)
                    {
                        $ex = $res[$i]->x;
                        $ey = $res[$i]->y;
                        imageline (self::$img, $ox, $oy, $ex, $ey, 0xFFFFFF);
                        $ox = $ex; $oy = $ey;
                    }
                    for ($i = 0 ; $i != count ($res) ; $i++)
                    {
                        imagesetpixel (self::$img, $res[$i]->x, $res[$i]->y, 0xFF);
                    }
                    imagepng (self::$img, self::$output_name);
printf (count($present)." nodes, ".round(memory_get_usage(true)/1024)."K\n");
printf ((count($res)-1)." moves\n");
                    return;
                }
                $signature = "$npx $npy $nvx $nvy";
                if (isset ($present[$signature])) continue;
                $border[] = new Node ($npx, $npy, $nvx, $nvy, $node);
                $present[$signature] = 1;
            }
        }
    }
}

ini_set("memory_limit","1000M");
Map::init ($argv[1]);
Map::solve();
?>

Language choice

PHP is pretty good at handling images.
It also has native associative memory, which makes programming BFS node lookup a breeze.

The downside is that the node identifiers hashing is not terribly time-efficient, so the result is anything but fast.

From my experiments with the previous racing challenge, I have no doubt C++11 and its hash tables would perform way better, but the source code would at the very least double, plus the need for whatever external png library (even LodePNG) would make for a messy build.

Perl and its more advanced offsprings would probably allow for a more compact and efficient code (due to better hashing performances), but I am not familiar enough with any of these to try a port.

BFS core

The search operates on a position+speed space, i.e. a node represents a given location visited with a given speed.
This of course makes for a pretty huge search space, but yields optimal results provided all possible track locations are examined.

Clearly, given the number of pixels on even a small image, an exhaustive search is out of the question.

Cuts

I chose to cut through the position space, by selecting only a subset of track pixels.
The solver will consider all positions within reach, limited only by acceleration.

The track is tiled with squares, whose maximal size is computed so that two adjacent squares can be reached with the maximal allowed acceleration (i.e. 14x14 pixels with the current speed limit).
After packing the track with big squares, increasingly smaller tiles are used to fill the remaining space.
Only the center of each tile is considered as a possible destination.

Here is an example:

track tiling example

This heuristic choice is enough to make the solver fail on the nightmare map. I suppose one could try to reduce the max tile size until some solution is found, but with the current settings the solver runs for something like an hour and uses 600 Mb, so more accurate results would require an unreasonable amount of time and memory.

As a second cut, squares of only 1 pixel might be left out.
This will of course degrade the solution or even prevent the solver from finding any, but it improves computation time a lot and usually yields pretty close results on "simple" maps.

For instance, on city, leaving out 1x1 pixel squares reduces the BFS tree nodes explored from 660K to about 90K for 47 vs 53 moves solutions.

BFS vs A*

A* needs more code and is not even guaranteed to produce faster results in the position/speed space, since evaluating the best next candidate is nothing as simple as in the classic position only space (which can easily be defeated with purpose-made cul-de-sacs anyway).

Besides, though PHP has some priority queues of sort, that by the way support dynamic reordering contray to their C++ cousins, I doubt they would be enough for an efficient A* implementation, and rewriting a binary heap or whatever dedicated A* queue structure would require way too many lines of code.

Wall check

I did not use a Bresenham algorithm to check for wall collisions, so the trajectory might clip the odd wall pixel. It should not allow to cross through a wall, though.

Performances

This solver is sure no six-legged jackrabbit.
Without the extra cut, a map can take over 10 minutes to solve on my middle-range PC.
I suggest setting tile minimal size to 2 or 3 if you want to fiddle with the code without spending ages waiting for a result.

Memory consumption is reasonable for this kind of algorithm and language: around 100 Mb or less except for nightmare that peaks above 600 Mb.

Results and scoring

Scores are given without the "minimal tile size" cut.

As a careful reader may deduce from my general comments, I don't care much about the golfing part of this challenge. I don't see how running my code through an obfuscator or deleting some optimizations and/or debug routines to shrink the source would make this any more fun.

Let just S be the source byte size for now :

track S+1020

track result

city S+470

city result

obstacles S+280

enter image description here

nightmare -> fails

user16991

Posted 2014-12-06T07:27:26.907

Reputation:

A very full and descriptive answer. It is very interesting to see your speed/space/complexity trade-offs and your thinking behind a language choice. – Logic Knight – 2014-12-10T01:13:29.727

Your straight line algorithm looks fine, however your car might have suffered some damage on the city track when it clipped some walls (eg: segment 7 and 14). You might have a small bug in your program. – Logic Knight – 2014-12-10T01:31:18.140

Well as I said I use a quick & dirty line check that does not match the Bresenham algorithm used by native line drawing. That explains the occasional 1 pixel difference that makes the car clip the walls, but still the check does its job of insuring the car won't drive straight across a wall, even 1 pixel wide. I could do a more precise coordinates computation if you think that's a fatal flaw, though. – None – 2014-12-10T01:54:31.343

From my point of view, a clipped wall here or there is not a problem. If a fellow competitor is concerned, we can discuss it. However, if it was my code, such an OBOB (out by one bug) would drive me crazy. – Logic Knight – 2014-12-10T02:35:28.760

1This would do the same to me if this pighead was able to crack the nightmare map. As long as it doesn't, it is not worthy of having its line drawing groomed :). – None – 2014-12-10T08:09:16.780

9

SecondRacer Java (1788 + 72*10 = 2508) (2708) (2887) (3088) (3382) (4109 + 72*10 = 4839) (4290 + 86*10 = 5150)

Similar to FirstRacer. But different in steps 2.2 and 3. which results in driving farsighted and using the gear.

  1. A* (A star)
  • 2.1 Build the A* search landscape.
  • 2.2. Find the best cell in *sight* and take the Euclidean distance into account. (Still looking only in directions N,S,E,W,NE,NW,SE,SW.) As a result SecondRacer finds a path with much less way points.
  1. Optimization is much elaborate now. The idea is to partition the given lines between two way points into as less as possible turns, while not violating the acceleration constraint.

Performance

None of these tracks is a problem for A*. Just some very few seconds (<10) to solve (even the "Nightmare Track: The Gauntlet") on my middle-range PC.

Path Style

I call it octopussy. By far not as elegant as the pighead solution (from kuroi neko).

Code Style

I entered into a moderate fighting mode keeping readability. But added a golfed version.

import java.awt.*;
import java.awt.image.*;
import java.io.*;
import java.util.*;
import javax.imageio.*;
import javax.swing.*;
import static java.lang.Math.*;

class AStar2 {
    BufferedImage img;
    int Width;
    int Height;
    int[][] cost;
    int best=Integer.MAX_VALUE;
    Point pBest;

    public static void main(String[] args) throws IOException {
        new AStar2().exec(args);
    }

    void exec(String[] args) throws IOException {
        img = ImageIO.read(new File(args[0]));
        Width=img.getWidth();
        Height=img.getHeight();

        draw(optimize(astar()));

        JFrame frame = new JFrame() {
            public void paint(Graphics g) {
                setSize(Width+17, Height+30+10);
                g.drawImage(img,8,30,null);
            }
        };
        frame.setVisible(true);
    }

    Vector<Point> astar() {
        Vector<Point> v0=findStart();
        while(v0.size()>0) v0=next(v0);

        for(Point p0 = pBest; p0!=null; p0=trackBack(p0)) v0.add(p0);
        return v0;
    }

    Vector<Point> findStart() {
        cost=new int[Width][Height];
        Vector<Point> v=new Vector<Point>();
        for (int i = 0; i < Width; i++)
            for (int j = 0; j < Height; j++) {
                if (getInfo(i,j)==1) {
                    cost[i][j]=0;
                    v.add(new Point(i, j));
                } else {
                    cost[i][j]=Integer.MAX_VALUE;
                    pBest=new Point(i, j);
                }
            }
        return v;
    }

    Vector<Point> next(Vector<Point> v0) {
        Vector<Point> v1=new Vector<Point>();
        for (Point p0 : v0) {
            int x=p0.x, y=p0.y,x1,y1,i1,c1, c0=cost[x][y];
            for (Point p : n(new Point(x,y))) {
                x1 = p.x; y1 = p.y;
                i1=getInfo(x1, y1);
                if (i1/2==1) {
                    c1=c0+(x1==x||y1==y?10:14);
                    if (c1<cost[x1][y1]) {
                        cost[x1][y1]=c1;
                        Point p1=new Point(x1, y1);
                        v1.add(p1);
                        if (i1==3) {
                            if (best>c1) {
                                best=c1;
                                pBest=p1;
                            }
                        }
                    }
                }
            }
        }
        return v1;
    }

    Point trackBack(Point p0) {
        Point p1=null, t;
        int x=p0.x, y=p0.y, i;
        double c0=0, c;
        for (Point p : n(new Point(0,0))) {
            for (i = 1; getInfo((t= new Point(x+i*p.x, y+i*p.y)).x, t.y)>0; i++) {
                c=cost[t.x][t.y]-cost[x][y]+5*sqrt((x-t.x)*(x-t.x) + (y-t.y)*(y-t.y));
                if (c0>c) {
                    c0=c;
                    p1= t;
                }
            }
        }
        return p1;
    }

    Vector<Point> n(Point p) {
        int [] c=new int[] {0, -1,-1, -1,-1, 0,-1, 1,0,1,1, 1,1, 0,1, -1};
        Vector<Point> v=new Vector<Point>();
        for (int i = 0; i < c.length; i+=2) v.add(new Point(p.x+c[i], p.y+c[i+1]));
        return v;
    }

    Vector<Point> optimize(Vector<Point> track) {
        Vector<Point> opt=new Vector<Point>();
        Point p0 = track.get(0);
        opt.add(p0);
        for (int i = 1; i < track.size(); i++) segmentAcceleration(opt, track.get(i));
        return opt;
    }

    boolean constraint(Point p0, Point p1, Point p2) {
        return abs(p2.x-p1.x-p1.x+p0.x) + abs(p2.y-p1.y-p1.y+p0.y) <= 15;
    }

    void segmentAcceleration(Vector<Point> opt, Point p1 ) {
        Point p0 = opt.lastElement();
        int d=max(abs(p0.x-p1.x), abs(p0.y-p1.y)), x=(p1.x-p0.x)/d, y=(p1.y-p0.y)/d, start=opt.size(),i;
        for (i = 0; i <=d; i++) opt.add(new Point(p0.x+x*i, p0.y+y*i));

        for(int success=1; success==1;) {
            success=0;
            for (int j = start; j < opt.size()-1; j++) {
                Point q=new Point(opt.get(j).x+x, opt.get(j).y+y);
                if (opt.get(j).x==opt.get(j+1).x && opt.get(j).y==opt.get(j+1).y) {
                    opt.remove(j);
                    success=1;
                } else if (j>1&&constraint(opt.get(j-2), opt.get(j-1), q) && constraint(opt.get(j-1), q, opt.get(j+1)) && (j>opt.size()-3 || constraint(q, opt.get(j+1), opt.get(j+2)))) {
                    opt.set(j, q);
                    success=1;
                }
            }
        }
    }

    void draw(Vector<Point> track) {
        Graphics2D g = img.createGraphics();
        Point p0=track.get(0);
        for (Point p1: track) {
            g.setColor(Color.WHITE);
            g.drawLine(p0.x, p0.y, p1.x, p1.y);
            img.setRGB(p0.x, p0.y, 0xff0000ff);
            img.setRGB(p1.x, p1.y, 0xff0000ff);
            p0=p1;
        }
    }

    int getInfo(int x, int y) {
        if (x<0 || x>=Width || y<0 || y>=Height) return 0;
        int rgb = img.getRGB(x, y), r=0xff&(rgb>>16), g=0xff&(rgb>> 8), b=0xff&(rgb>> 0);
        switch (rgb) {
        case 0xffffff00: return 1;
        case 0xff000000: return 3;
        default: if (30<=r&&r<=220&&r==g&&g==b) return 2;
        }
        return 0;
    }
}

golfed -> WARNING: it does in place replacement of the original file!

import java.awt.*;import javax.imageio.*;import static java.lang.Math.*;class
A{class B{int C,D;}class E extends java.util.Vector<B>{};static java.awt.image.BufferedImage
F;int G=F.getWidth(),H=F.getHeight(),I[][]=new int[G][H],J=-1>>>1,K,L,M=255,N;B
O,P,Q;public static void main(String[]R)throws Exception{F=ImageIO.read(new
java.io.File(R[0]));new A().S();ImageIO.write(F,"PNG",new java.io.File(R[0]));}void
S(){E U=new E(),V=new E();for(K=0;K<G;K++)for(L=0;L<H;L++)if(W(K,L)==1)U.add(X(K,L));else
I[K][L]=J;while(U.size()>0){P=U.remove(0);int C=P.C,D=P.D,Y,Z,a,b;for(N=0;N<9;N++)if(N!=4)if((a=W(Y=C+N/3-1,Z=D+N%3-1))>0&&I[Y][Z]>(b=I[C][D]+(Y==C||Z==D?10:14))){I[Y][Z]=b;U.add(X(Y,Z));if(a==3&&J>b){J=b;O=Q;}}}P=O;while(O!=null){U.add(O);J=O.C;L=O.D;double
c=0,d;O=null;for(N=0;N<9;N++)if(N!=4)for(K=1;W(X(J+K*(N/3-1),L+K*(N%3-1)).C,Q.D)>0;K++)if(c>(d=I[Q.C][Q.D]-I[J][L]+5*sqrt((J-Q.C)*(J-Q.C)+(L-Q.D)*(L-Q.D)))){c=d;O=Q;}}Graphics2D
e=F.createGraphics();V.add(P);for(K=1;K<U.size();K++){O=P;P=U.get(K);e.setColor(Color.WHITE);e.drawLine(O.C,O.D,P.C,P.D);int
f=max(abs(O.C-P.C),abs(O.D-P.D)),C=(P.C-O.C)/f,D=(P.D-O.D)/f,start=V.size();for(L=0;L<=f;L++)V.add(X(O.C+L*C,O.D+L*D));while(f>0)for(f=0,L=start;L<V.size()-1;L++)if(V.get(L).C==V.get(L+1).C&&V.get(L).D==V.get(L+1).D)V.remove(L);else
if(L>1&&g(V.get(L-2),V.get(L-1),X(V.get(L).C+C,V.get(L).D+D))&&g(V.get(L-1),Q,V.get(L+1))&&(L>V.size()-3||g(Q,V.get(L+1),V.get(L+2)))){V.set(L,Q);f=1;}}for(B
h:V)F.setRGB(h.C,h.D,~0xffff00);}B X(int C,int D){Q=new B();Q.C=C;Q.D=D;return
Q;}boolean g(B O,B P,B i){return abs(i.C-P.C-P.C+O.C)+abs(i.D-P.D-P.D+O.D)<16;}int
W(int C,int D){if(C>=0&&C<G&&D>=0&&D<H){int j=F.getRGB(C,D),Q=j>>16&M,e=j>>8&M;if(j==~M)return
1;if(j==M<<24)return 3;if(30<=Q&&Q<=220&&Q==e&&e==(M&j))return 2;}return
0;}}

All images are shown on their A* gradient landscape. Starting from light yellow to brown (=dark yellow). While - according to A* - the resulting path is tracked backwards (here from brown to light yellow).

Race Track S+175 Race Track

Obstacle Course S+47 Obstacle Course

City S+72 City

Gauntlet S+1133 Gauntlet

Bob Genom

Posted 2014-12-06T07:27:26.907

Reputation: 846

Well done Bob. Golfing is not easy in Java. It's like you have a golf-bag full of putters ;-) – Logic Knight – 2015-01-04T13:04:06.407

@CarpetPython: You are right. So, I expect nimi to catch up soon :-) – Bob Genom – 2015-01-04T13:15:57.063

@BobGenom: Haskell is also not easy - at least for me. This is my very first golf challange. I've already found some further bytes to save ... will update later. – nimi – 2015-01-04T16:42:44.130

@nimi: I'm very close to running out of steam. My downsizing steps getting smaller. (I wish my car would be faster.) – Bob Genom – 2015-01-04T21:09:55.100

@BobGenom: Me too. I think, I'm close to the limit. – nimi – 2015-01-04T22:07:11.387

9

Tutu - Haskell - (3460 2654 2476 2221 2060 1992 1900 + 50x10 = 2400)

Strategy:

  1. find A*
  2. bloat the path with it’s neighbors (distance 2)
  3. find again A*, but this time in the position+speed space, just like the pighead racer

Golf:

  • left out most of the error checking, so the program assumes that there are always start and end points in the map and a path inbetween.
  • short variable names

I’m not a Haskell golfer, so I don’t know how much can be further saved (Edit: turned out to be quite a bunch).

Performance:

The Gauntlet runs in 9:21min on my 1,7GHz Core i5 from 2011. The City takes 7.2sec. (used -O1 with ghc, -O2 makes the program awfully slow)

Tweaking options are the thickness of the bloated path. For the smaller maps distance 3 saves one or two steps, but The Gauntlet runs too long, so I stay with distance 2. The race starts always at the first (i.e. top left) yellow pixel, optimizing by hand should save an additional step.

Rule conformity:

„You can't use path-finding libraries.“ - I do, but the source is included. The A* search functions are slightly heavily golfed versions of Cale Gibbard’s Data.Graph.AStar library (see http://hackage.haskell.org/package/astar).

The City - 50 steps The City - 50 steps

The Gauntlet - 722 steps The Gauntlet - 722 steps

Ungolfed:

import System.Environment
import Data.Maybe (fromJust)
import Graphics.GD
import qualified Data.Matrix as M
import qualified Data.List as L
import qualified Data.Set as S
import qualified Data.Map as Map
import qualified Data.PSQueue as PSQ

main = do
    trPNG <- loadPngFile =<< fmap head getArgs
    (sX, sY) <- imageSize trPNG
    px <- mapM (flip getPixel trPNG) [(x,y) | y <- [0..sY-1],x <- [0..sX-1]]
    let tr = M.fromList sY sX (map (rgbaToTok . toRGBA) px)
    let rt = findRt tr
    let vrt = findVRt (head rt) (last rt) (bloat rt tr) tr
    let wayPt = map ((\(a,b)->(b-1,a-1)) . fst) vrt
    mapM (\(p1,p2) -> drawLine p1 p2 (rgb 255 255 255) trPNG
        >> setPixel p1 (rgb 0 0 255) trPNG) (zip wayPt (tail wayPt))
    savePngFile "out1.png" trPNG 
    print $ length vrt - 1

findVRt p1 p2 rt tr = (p1, (0,0)) : fromJust (aStar nghb (\_ _ -> 100)
        (\(pos,_) -> fromJust $ Map.lookup pos rt)
        ((==) p2 . fst) (p1, (0,0)))
    where
    nghb ((y,x), (vy,vx)) =
        S.fromList [(newp, (vy+dy,vx+dx)) |
            dy <- [-15 .. 15],
            let ady = abs dy,
            dx <- [-15+ady .. 15-ady],
            not $ dy==0 && dx == 0 && vy == 0 && vx == 0,
            let newp = (y+vy+dy,x+vx+dx),
            Map.member newp rt,
            all ((/=) 1 . (M.!) tr) (bresenham (y,x) newp)]

bloat rt tr = foldr (\(p,h) -> Map.insert p h) Map.empty
                (zip (reverse $ f $ f rt) [0..])
    where
    f = concatMap (n8 tr)

rgbaToTok (r, g, b, _)
    | r+g+b == 0 = 3
    | r==255 && g==255 && b==0 = 2
    | r==g && r==b && 30 <= r && r <= 220 = 0
    | otherwise = 1

findRt tr = s : fromJust (aStar nghb cost (const 1) ((==) 3 . (M.!) tr) s)
    where
    cost (y1,x1) (y2,x2) = if (x1==x2 || y1==y2) then 408 else 577
    nghb = S.fromList . n8 tr
    s = head [(y,x) | y <- [1..M.nrows tr], x <- [1..M.ncols tr],
            M.getElem y x tr == 2]

n8 tr p@(y,x) = filter ((/=) 1 . (M.!) tr) (n8' y x)
    where
    n8' y x | y==1 || x==1 || y == M.nrows tr || x == M.ncols tr = [p]
        | otherwise = [ (y-1,x-1), (y-1,x), (y-1,x+1), (y,x-1),
                (y,x+1), (y+1,x-1), (y+1,x), (y+1,x+1) ]

bresenham start@(y0,x0) end@(y1,x1) = walk start (el `div` 2)
    where
    walk p@(y,x) err
        | p == end = [p]
        | err-es < 0 = p : walk (y+sdy,x+sdx) (err-es+el)
        | otherwise = p : walk (y+pdy,x+pdx) (err-es)

    dx = x1-x0; dy = y1-y0;
    adx = abs dx; ady = abs dy
    sdx = signum dx; sdy = signum dy
    (pdx,pdy,es,el) = if adx > ady then (sdx,0,ady,adx) else (0,sdy,adx,ady)

data AStar a c = AStar {
    vi :: !(S.Set a), wa :: !(PSQ.PSQ a c), sc :: !(Map.Map a c),
    mH :: !(Map.Map a c), ca :: !(Map.Map a a), en :: !(Maybe a) }
    deriving Show

aStarInit s = AStar S.empty (PSQ.singleton s 0) (Map.singleton s 0)
    Map.empty Map.empty Nothing

aStar graph dist heur goal start =
    let s = runAStar graph dist heur goal start
    in case en s of
        Nothing -> Nothing
        Just e -> Just (reverse . takeWhile (not . (== start))
                    . iterate (ca s Map.!) $ e)

runAStar graph dist heur goal start = aStar' (aStarInit start)
    where
    aStar' s = case PSQ.minView (wa s) of
        Nothing -> s
        Just (x PSQ.:-> _, w') ->
            if goal x
            then s { en = Just x }
            else aStar' $ L.foldl' (expand x) (s { wa = w',
                    vi = S.insert x (vi s)})
                    (S.toList (graph x S.\\ vi s))
    expand x s y =
        let vi = sc s Map.! x + dist x y
        in case PSQ.lookup y (wa s) of
            Nothing -> link x y vi (s { mH
                    = Map.insert y (heur y) (mH s) })
            Just _ -> if vi<sc s Map.! y then link x y vi s else s
    link x y v s = s {
            ca = Map.insert y x (ca s),
            sc = Map.insert y v (sc s),
            wa = PSQ.insert y (v + mH s Map.! y) (wa s) }

Golfed:

import System.Environment;import Graphics.GD;import Data.Matrix;import qualified Data.Set as S;import qualified Data.Map as J;import qualified Data.PSQueue as Q
j(Just x)=x;e(y,x)=(x-1,y-1);u=signum;q=J.empty;m=reverse;n=Nothing;z=255;s=abs;t=1<2;f(a,b)(c,d)|b==d||a==c=2|t=3;rT(r,g,b,_)|r+g+b==0=3|r==z&&g==z&&b==0=2|r==g&&r==b&&30<=r&&r<=220=0|t=5
main=do
 i:_<-getArgs;t<-loadPngFile i;(a,b)<-imageSize t;p<-mapM(flip getPixel t)[(x,y)|y<-[0..b-1],x<-[0..a-1]];let r=fromList b a$map(rT.toRGBA)p;s:_=[(y,x)|y<-[1..b],x<-[1..a],getElem y x r==2];c p@(y,x)=filter((<5).(!)r)$if y==1||x==1||y==b||x==a then[p]else[(a,b)|a<-[y-1..y+1],b<-[x-1..x+1],a/=y||b/=x];y=s:j(aS(S.fromList.c)f(\_->1)((==3).(!)r)s);l=concatMap c;w=map(e.fst)$fV(head y)(last y)(foldr(\(p,h)->J.insert p h)q$zip(m$l$l y)[0..])r
 mapM(\(c,d)->drawLine c d(rgb z z z)t>>setPixel c(rgb 0 0 z)t)$zip w$tail w;savePngFile"o.png"t
fV c d r t=(c,(0,0)):j(aS l(\_ _->99)(\(q,_)->j$J.lookup q r)((==d).fst)(c,(0,0)))where l((y,x),(a,b))=S.fromList[(w,(a+c,b+d))|c<-[-15..15],d<-[s c-15..15-s c],any(/=0)[a,b,c,d],let w=(y+a+c,x+b+d),J.member w r,all((<5).(!)t)$br(y,x)w]
br q@(i,j)r@(k,l)=w q$f`div`2where w p@(y,x)e|p==r=[p]|e-o<0=p:w(y+g,x+h)(e-o+f)|t=p:w(y+m,x+n)(e-o);a=s$l-j;b=s$k-i;h=u$l-j;g=u$k-i;(n,m,o,f)|a>b=(h,0,b,a)|t=(0,g,a,b)
data A a c=A{v::S.Set a,w::Q.PSQ a c,k::J.Map a c,mH::J.Map a c,ca::J.Map a a,en::Maybe a}deriving Show
aS g d h o u=b$en s where b Nothing=n;b(Just e)=Just(m.takeWhile(/=u).iterate(ca s J.!)$e);s=l$A S.empty(Q.singleton u 0)(J.singleton u 0)q q n;i x y v s=s{ca=J.insert y x$ca s,k=J.insert y v$k s,w=Q.insert y(v+mH s J.!y)$w s};l s=b$Q.minView$w s where b Nothing=s;b(Just(x Q.:->_,w'))|o x=s{en=Just x}|t=l$foldl(r x)(s{w=w',v=S.insert x$v s})$S.toList$g x S.\\v s;r x s y=b$Q.lookup y$w s where v=k s J.!x+d x y;b Nothing=i x y v$s{mH=J.insert y(h y)$mH s};b(Just _)|v<k s J.!y=i x y v s|t=s

Tutu siblings -TS#1 - (1764+43 = 2194)

Edit: TS#1 now separate answer.

Edit II: The path for The City is

[(6,7),(21,7),(49,5),(92,3),(126,4),(145,5),(149,6),(153,22),(163,47),(180,64),
(191,73),(191,86),(185,107),(177,122),(175,130),(187,137),(211,147),(237,162),
(254,171),(277,171),(299,175),(321,194),(345,220),(364,237),(370,252),(365,270),
(360,276),(368,284),(387,296),(414,315),(438,330),(463,331),(484,321),(491,311),
(498,316),(508,333),(524,354),(525,375),(519,404),(511,424),(508,434),(513,437),
(533,440),(559,444),(580,458),(591,468),(591,482),(591,511),(598,532),(605,539),
(606,537)]

In The Gauntlet Tutu moves as follows

[(99,143),(114,143),(137,150),(150,161),(149,173),(145,180),(141,197),(138,223),
(135,234),(143,241),(166,248),(186,250),(192,251),(192,261),(192,279),(195,285),
(209,287),(232,284),(248,273),(257,261),(272,256),(279,255),(284,245),(294,243),
(309,231),(330,226),(354,233),(380,253),(400,265),(421,271),(436,268),(438,266),
(440,269),(441,277),(450,278),(470,276),(477,276),(478,285),(481,307),(490,330),
(486,352),(471,370),(449,384),(435,391),(433,401),(446,411),(462,430),(464,450),
(459,477),(454,493),(457,514),(462,522),(472,523),(479,529),(491,531),(493,538),
(496,547),(503,546),(516,545),(519,549),(524,566),(531,575),(531,581),(535,576),
(538,557),(541,523),(545,475),(551,414),(559,342),(565,282),(570,236),(574,204),
(575,184),(574,177),(572,179),(568,174),(568,158),(569,144),(572,143),(578,154),
(585,160),(588,155),(593,140),(598,140),(605,153),(610,156),(611,170),(611,182),
(608,182),(598,175),(594,171),(590,176),(587,195),(589,224),(593,266),(599,321),
(605,376),(609,418),(612,446),(610,465),(615,478),(608,494),(605,521),(611,542),
(618,549),(622,551),(621,563),(611,572),(614,581),(623,581),(630,581),(630,573),
(636,556),(639,551),(642,531),(647,520),(640,511),(637,491),(639,461),(641,416),
(643,356),(647,289),(650,235),(652,195),(647,163),(645,143),(645,136),(653,136),
(670,138),(673,139),(676,155),(679,175),(681,181),(669,188),(662,194),(662,208),
(665,234),(669,274),(674,328),(681,395),(687,457),(692,505),(696,540),(700,560),
(703,566),(706,557),(707,535),(708,498),(711,448),(716,385),(720,325),(723,278),
(726,246),(729,229),(732,227),(733,238),(733,263),(733,303),(733,358),(733,428),
(733,483),(733,523),(732,549),(731,560),(728,558),(726,565),(726,575),(721,575),
(720,586),(720,592),(716,594),(715,608),(715,619),(711,619),(692,619),(658,619),
(609,619),(545,619),(466,619),(372,619),(285,619),(213,619),(155,619),(112,619),
(84,619),(70,618),(70,616),(70,599),(70,567),(70,520),(70,458),(70,381),
(70,300),(70,234),(70,183),(70,147),(70,126),(71,119),(80,119),(104,119),
(143,119),(197,119),(266,119),(350,119),(449,119),(563,119),(681,120),(784,121),
(873,121),(947,121),(1006,121),(1050,121),(1079,121),(1093,121),(1093,122),
(1086,131),(1069,145),(1059,151),(1040,151),(1006,151),(973,150),(955,149),
(950,150),(956,155),(977,160),(994,175),(1003,183),(1003,197),(993,214),
(987,220),(993,223),(1011,223),(1044,223),(1079,229),(1104,240),(1124,242),
(1134,239),(1134,231),(1134,221),(1139,218),(1156,218),(1177,217),(1183,216),
(1191,202),(1208,182),(1231,154),(1249,135),(1259,123),(1264,121),(1264,129),
(1264,152),(1264,190),(1264,243),(1264,311),(1264,393),(1264,460),(1264,512),
(1264,550),(1264,573),(1263,582),(1256,582),(1234,582),(1197,582),(1160,575),
(1132,562),(1118,548),(1113,538),(1107,541),(1099,549),(1102,561),(1113,570),
(1110,578),(1095,583),(1073,581),(1066,579),(1060,566),(1063,559),(1075,554),
(1072,549),(1065,542),(1051,539),(1043,528),(1023,520),(990,511),(970,500),
(953,501),(935,516),(911,534),(899,551),(891,573),(883,580),(867,581),(859,575),
(858,571),(843,566),(830,553),(832,540),(828,527),(819,520),(825,513),(839,506),
(842,495),(843,474),(844,468),(854,468),(877,467),(891,460),(895,452),(901,452),
(906,447),(909,443),(909,441),(915,435),(912,430),(914,429),(908,423),(904,421),
(899,418),(893,417),(879,409),(854,400),(842,390),(842,377),(839,362),(836,362),
(820,360),(812,352),(812,337),(812,307),(814,288),(815,282),(827,280),(834,284),
(850,282),(873,277),(889,280),(891,284),(891,301),(897,320),(903,324),(916,320),
(925,310),(935,314),(953,325),(967,337),(976,345),(981,346),(986,362),(999,378),
(1006,385),(1007,387),(1008,387),(1015,382),(1017,382),(1018,381),(1022,386),
(1021,401),(1008,413),(1009,425),(1014,426),(1031,425),(1038,429),(1047,425),
(1053,429),(1067,426),(1076,425),(1090,427),(1099,424),(1113,426),(1134,427),
(1147,431),(1150,430),(1152,437),(1147,438),(1128,438),(1105,443),(1093,450),
(1089,453),(1085,449),(1075,452),(1064,460),(1055,458),(1052,462),(1049,460),
(1042,464),(1025,463),(1015,463),(1010,470),(1013,471),(1021,472),(1027,476),
(1033,477),(1042,484),(1052,480),(1059,486),(1076,487),(1099,497),(1134,510),
(1169,523),(1191,535),(1205,540),(1210,539),(1210,528),(1210,502),(1210,461),
(1209,409),(1208,372),(1207,349),(1206,341),(1192,335),(1165,327),(1132,310),
(1084,293),(1045,273),(997,256),(961,240),(934,229),(922,218),(919,201),
(917,197),(906,199),(892,212),(876,212),(845,212),(809,212),(781,219),(768,226),
(768,235),(768,259),(768,298),(768,352),(768,421),(769,489),(769,543),(769,582),
(769,606),(769,615),(775,615),(796,615),(832,615),(883,615),(949,615),
(1030,615),(1110,615),(1175,615),(1225,615),(1261,614),(1282,613),(1288,612),
(1296,598),(1296,577),(1296,541),(1296,490),(1296,424),(1296,343),(1296,264),
(1296,200),(1296,151),(1296,116),(1296,96),(1295,90),(1285,90),(1260,90),
(1220,90),(1165,90),(1095,90),(1010,90),(920,90),(844,90),(783,90),(737,90),
(706,90),(690,90),(688,89),(689,86),(681,78),(671,82),(663,90),(648,90),
(618,90),(573,90),(517,90),(476,90),(450,90),(438,89),(439,86),(431,78),
(421,82),(413,90),(398,90),(381,88),(369,78),(357,83),(353,90),(341,90),
(314,90),(297,88),(287,78),(277,82),(269,90),(254,90),(224,90),(179,90),
(123,90),(82,90),(56,90),(43,92),(43,96),(43,115),(43,149),(43,198),(43,262),
(43,341),(43,428),(43,500),(43,557),(43,599),(44,627),(45,640),(49,641),
(67,641),(100,641),(148,641),(211,641),(289,641),(382,641),(490,641),(613,641),
(750,641),(872,641),(979,641),(1071,641),(1148,641),(1212,640),(1261,639),
(1295,638),(1315,636),(1321,633),(1321,621),(1321,594),(1321,552),(1321,495),
(1321,423),(1321,336),(1321,254),(1321,187),(1321,135),(1321,98),(1321,75),
(1320,66),(1313,66),(1291,66),(1254,66),(1207,67),(1175,68),(1157,68),(1154,68),
(1154,75),(1146,75),(1123,75),(1102,74),(1096,73),(1096,69),(1091,66),(1074,66),
(1042,66),(1007,66),(986,65),(980,64),(980,60),(975,57),(958,57),(926,57),
(891,58),(871,59),(866,60),(865,66),(855,66),(830,66),(790,66),(735,66),
(667,66),(614,66),(575,66),(550,65),(540,64),(540,60),(535,57),(518,57),
(489,58),(474,60),(474,62),(472,66),(459,66),(431,66),(388,66),(330,66),
(269,66),(223,66),(191,66),(174,66),(171,65),(168,56),(158,55),(150,61),
(149,66),(138,66),(112,66),(98,63),(95,57),(83,57),(65,59),(61,62),(59,66),
(46,66),(25,67),(18,69),(18,79),(18,104),(18,144),(18,199),(18,269),(18,354),
(18,441),(18,513),(18,570),(18,612),(18,639),(19,652),(26,656),(38,663),
(58,663),(93,663),(143,663),(208,663),(288,663),(383,663),(493,663),(618,663),
(758,663),(884,663),(995,663),(1091,663),(1172,663),(1239,663),(1291,663),
(1328,663),(1350,663),(1358,662),(1361,651),(1376,637),(1378,621),(1374,597),
(1378,574),(1378,541),(1375,519),(1383,501),(1376,483),(1370,478),(1370,464),
(1373,438),(1379,400),(1379,366),(1369,337),(1369,303),(1369,272),(1368,255),
(1382,238),(1381,221),(1371,209),(1375,196),(1380,170),(1374,143),(1367,129),
(1372,112),(1373,85),(1365,64),(1358,57),(1356,41),(1353,39),(1350,41),
(1346,37),(1336,36),(1333,32),(1317,30),(1288,30),(1244,30),(1185,30),(1141,30),
(1102,22),(1057,22),(1026,21),(1005,23),(993,21),(988,25),(975,22),(972,24),
(959,21),(943,24),(937,29),(920,30),(889,30),(843,30),(788,30),(747,30),
(706,39),(664,36),(629,38),(591,34),(559,34),(538,30),(506,30),(465,30),
(431,22),(391,23),(356,22),(328,23),(308,30),(280,30),(237,30),(179,30),
(106,30),(30,28)]

nimi

Posted 2014-12-06T07:27:26.907

Reputation: 34 639

This entry is in the running for the bounty, but you will need a valid path list to claim it. – Logic Knight – 2015-01-19T14:33:09.367

@CarpetPython: I've added the paths for City and Gauntlet. – nimi – 2015-01-19T16:54:23.890

City and Gauntlet paths pass acceleration check. – Logic Knight – 2015-01-19T17:07:33.207

any idea why -O2 slows the program?? weird. tried -O3? – proud haskeller – 2015-01-20T20:14:20.237

by the way, seems you're using Maybe a lot. maybe you can replace Maybe with lists: Maybe a is [a], Nothing is [] and Just x is [x]. the Maybe monad becomes equivalent to the List monad. you can then use a lot of list functions for them: null, head, (:[]), map and so forth. – proud haskeller – 2015-01-20T20:17:33.273

@proudhaskeller: never thought about replacing the Maybe with lists. Maybe I'll give it a try (but then in the TS#1 racer, where I have already removed some of the Maybes). -- No idea why -O2 is so slow. – nimi – 2015-01-21T20:23:13.780

5

Star crossed racer - PHP - 4083+440 = way too heavy

Ok, after 3 weeks without an Internet connection (that's what happens when one of the most ferocious competitor of your provider happens to be in charge of the building patch bay, or so it does in Paris, France at least) I can at last publish my next attempt.

This time I used the A* algorithm, and a more efficient waypoint distribution strategy.
As an added bonus, I wrote some kind of PHP cruncher to address the golfing part of the challenge.

And now that the solver works on all proposed maps, the line tracing bug has been fixed.
No more wall clipping (though wall grazing still occurs, as it should :)).

running the code

give the code whatever name you fancy (runner.php for instance), then invoke it like so:

php runner.php track.png

After sitting silent for quite a while, it should produce a _track.png output showing the solution.

As you can see on the output pictures, the code is really slow. You've been warned.

Of course my own private version is full of traces and produces nice representations of various informations (including a periodic picture showing the A* progress to help killing time), but there is a price to pay for golfing...

golfed code

<?php
define("_",15);define("a",1e3);class _{function _($a=0,$_=0){$this->a=$a;$this->_=$_;}function b(){return sqrt($this->a*$this->a+$this->_*$this->_);}function a(){$_=$this->b();if($_==0)$_=1;return new _($this->a/$_,$this->_/$_);}}class c{static$_=0;function c($_,$a,$b){$this->c=$b;$this->a=++c::$_;$this->_=new _($_,$a);a::$a[$_][$a]=$this;}function _($_){return(isset($this->b[$_->a]))?$this->b[$_->a]:$this->b[$_->a]=$_->b[$this->a]=a($_->_->a,$_->_->_,$this->_->a,$this->_->_);}}define("c",8/_);define("b",2/a);class d{function d($a,$b,$c=0,$d=0,$_=null){$this->a=$a;$this->_=$b;$this->f=$c;$this->e=$d;$this->d=$_;$this->b=$_==null?0:$_->b+a;$this->c=floor((sqrt(1+c*abs(a::$_[$a][$b]))-1)/b)-a;}}function a($c,$b,$g,$f){$e=$g-$c;$d=$f-$b;$_=2*max(abs($e),abs($d));if($_==0)return 1;$c+=.5;$b+=.5;for($a=1;$a<=$_;$a++)if(!isset(a::$_[$c+$a/$_*$e][$b+$a/$_*$d]))return 0;return 1;}class b{static$a,$_;function _($l,$_,$k){$g=log(.5)/log($l/$_);for($a=-$_;$a<=$_;$a++)for($b=-$_;$b<=$_;$b++){$d=sqrt($a*$a+$b*$b);if($d>=$_)continue;$j=pow(sin(M_PI*pow($d/$_,$g)),$k);$c=new _($a,$b);$i=$c->a();$c->b=$d;$c->d=$j*$i->a;$c->c=$j*$i->_;$h[]=$c;}usort($h,function($b,$a){$_=$b->b-$a->b;return($_>0)?1:(($_<0)?-1:0);});foreach($h as$e)b::$a[$e->b][]=$e;for($a=-$_;$a<=$_;$a++)for($b=-$_;$b<=$_;$b++){$e=new _($a,$b);$d=sqrt($a*$a+$b*$b);for($f=$_;$f>=$d;$f--)b::$_[$f][]=$e;}foreach(b::$_ as$g=>$m)usort(b::$_[$g],function($b,$a){$_=$b->b()-$a->b();return($_<0)?-1:(($_>0)?1:0);});}}b::_(2.5,6,8);class a{static$a,$_;static function _($S){$k=imagecreatefrompng($S);for($a=0;$a!=imagesx($k);$a++)for($_=0;$_!=imagesy($k);$_++){$n=0;$o=imagecolorat($k,$a,$_);if($o==0){$d_[]=new _($a,$_);$n=1;}else if($o==0xFFFF00){$e_[]=new _($a,$_);$n=2;}else{$m=($o>>16)&0xFF;$c_=($o>>8)&0xFF;$a_=$o&0xFF;if($m==$c_&&$m==$a_&&$m>=30&&$m<=220)$n=3;}if($n){$Z[$a][$_]=1;if($n!=3)$b_[]=new c($a,$_,$n);}}for($a=-_;$a<=_;$a++)for($_=-_;$_<=_;$_++)if(abs($a)+abs($_)<=_)$f_[]=new _($a,$_);$l_=array(new _(-1,0),new _(1,0),new _(0,-1),new _(0,1));foreach($d_ as$v){$c[]=new _($v->a,$v->_);a::$_[$v->a][$v->_]=0;}while(count($c)){$t=array_shift($c);$g_=a::$_[$t->a][$t->_]+1;foreach($l_ as$e){$f=$t->a+$e->a;$j=$t->_+$e->_;if(!isset($Z[$f][$j]))continue;if(isset(a::$_[$f][$j]))continue;a::$_[$f][$j]=$g_;$c[]=new _($f,$j);}}foreach(a::$_ as$a=>$g)foreach($g as$_=>$q){$i=0;$E=$H=0;foreach(b::$a as$n_=>$J){foreach($J as$b){if(!isset(a::$_[$a+$b->a][$_+$b->_])){$E+=$b->d;$H+=$b->c;$i++;}}if($i!=0){$E/=$i;$H/=$i;break;}}$W[$a][$_]=new _($E,$H);}foreach(a::$_ as$a=>$g)foreach($g as$_=>$q){$T=$W[$a][$_];$u=$T->a();$R=0;$i=0;foreach(b::$_[1]as$e){@$b=$W[$a+$e->a][$_+$e->_];if(!$b)continue;$V=$b->a();$d=$e->a();$R-=($u->a*$V->_-$u->_*$V->a)*($u->a*$d->_-$u->_*$d->a);$i++;}$p[$a][$_]=(12*$R/$i+1)*$T->b();}$m_=1;$Y=6;$x=0;foreach($p as$a=>$g)foreach($g as$_=>$q)$x=max($x,$q);$h_=($m_-$Y)/$x;foreach($p as$a=>$g)foreach($g as$_=>$q)$X[($Y+$h_*max($q,0))*1e5][]=new _($a,$_);ksort($X);foreach($X as$m=>$J)foreach($J as$b){$a=$b->a;$_=$b->_;if(!isset($p[$a][$_]))continue;$b_[]=new c($a,$_,3);unset($p[$a][$_]);$k_=0;foreach(b::$_[$m/1e5]as$U){$f=$a+$U->a;$j=$_+$U->_;if(a($a,$_,$f,$j))unset($p[$f][$j]);else if(++$k_==2)break;}}foreach($e_ as$s){$e=new d($s->a,$s->_);$c[$e->b+$e->c][]=$e;$y[$s->a." ".$s->_." 0 0"]=$e;}ksort($c);while(count($c)){reset($c);$z=key($c);$r=array_shift($c[$z]);if(empty($c[$z]))unset($c[$z]);$A=$r->a;$C=$r->_;$M=$r->f;$O=$r->e;$i_=a::$a[$A][$C];$l="$A $C $M $O";unset($y[$l]);$j_[$l]=1;foreach($f_ as$P){$B=$M+$P->a;$D=$O+$P->_;$G=$A+$B;$F=$C+$D;@$I=a::$a[$G][$F];if(!$I)continue;$l="$G $F $B $D";if(@$j_[$l]||@$y[$l])continue;if(!$I->_($i_))continue;$d=new d($G,$F,$B,$D,$r);$b=$d->b+$d->c;$__=!isset($c[$b]);$c[$b][]=$d;if($__)ksort($c);$y[$l]=$d;if($I->c==1){for($h=array();$d!=null;$d=$d->d)array_unshift($h,$d);$N=$h[0]->a;$K=$h[0]->_;for($w=1;$w!=count($h);$w++){$L=$h[$w]->a;$Q=$h[$w]->_;imageline($k,$N,$K,$L,$Q,0xFFFFFF);$N=$L;$K=$Q;}foreach($h as$b)imagesetpixel($k,$b->a,$b->_,0xFF);imagepng($k,"_".$S);return;}}}}}ini_set("memory_limit","3G");a::_($argv[1]);

ungolfed version

<?php
define ("ACCEL_MAX", 15);   // maximal acceleration value
define ("MOVE_UNIT", 1000); // heuristic distance to goal precision

class Point {
    function __construct ($x=0, $y=0)
    {
        $this->x = $x;
        $this->y = $y;
    }

    function norm () { return sqrt ($this->x*$this->x + $this->y*$this->y); }

    function normalized() { $n=$this->norm(); if ($n == 0) $n=1; return new Point ($this->x / $n, $this->y / $n); }
}

class Waypoint {
    static $id = 0;

    function __construct ($x, $y, $type)
    {
        // create waypoint
        $this->type = $type;
        $this->id = ++self::$id;
        $this->center = new Point ($x, $y);

        // update waypoint lookup map
        Map::$waypoint_lookup[$x][$y] = $this;
    }

    function can_reach ($target)
    {
        return (isset($this->reachable[$target->id])) 
            ? $this->reachable[$target->id]
            : $this->reachable[$target->id] = $target->reachable[$this->id] = on_track ($target->center->x, $target->center->y, $this->center->x, $this->center->y);
    }
}

define ("L", 8/ACCEL_MAX);
define ("M", 2/MOVE_UNIT);
class Node {
    function __construct ($x, $y, $speedx=0, $speedy=0, $parent=null)
    {
        $this->x = $x;
        $this->y = $y;
        $this->speedx = $speedx;
        $this->speedy = $speedy;
        $this->parent = $parent;

        // previous moves
        $this->moves = $parent == null ? 0 : $parent->moves+MOVE_UNIT;

        // remaining moves heuristic estimation
        $this->dist = floor((sqrt (1+L*abs(Map::$dist_to_goal[$x][$y])) - 1)/M) - MOVE_UNIT;
    }
}

function on_track ($ox, $oy, $ex, $ey)
{
    $sx = $ex - $ox;
    $sy = $ey - $oy;
    $range = 2*max (abs ($sx), abs ($sy));
    if ($range == 0) return 1;
    $ox+=.5;
    $oy+=.5;
    for ($s = 1 ; $s <= $range ; $s++) if (!isset (Map::$dist_to_goal[$ox + $s/$range*$sx][$oy + $s/$range*$sy])) return 0;
    return 1;
}

class Border {
    static $circle, $area;

    function init ($dopt, $dmax, $erf)
    {
        $k = log (.5)/log($dopt/$dmax);

        for ($x = -$dmax ; $x <= $dmax ; $x++)
        for ($y = -$dmax ; $y <= $dmax ; $y++)
        {
            $d = sqrt ($x*$x+$y*$y);
            if ($d >= $dmax) continue;
            $i = pow(sin (M_PI*pow($d/$dmax,$k)),$erf); // a function that will produce a kind of asymetric gaussian
            $pt = new Point($x,$y);
            $pn = $pt->normalized();
            $pt->d = $d;
            $pt->ix = $i * $pn->x;
            $pt->iy = $i * $pn->y;
            $points[] = $pt;
        }
        usort ($points, function ($a,$b) { $d = $a->d - $b->d; return ($d > 0) ? 1 : (($d < 0) ? -1 : 0); });
        foreach ($points as $p) self::$circle[$p->d][] = $p;

        for ($x = -$dmax ; $x <= $dmax ; $x++)
        for ($y = -$dmax ; $y <= $dmax ; $y++)
        {
            $p = new Point ($x, $y);
            $d = sqrt ($x*$x+$y*$y);
            for ($r = $dmax ; $r >= $d ; $r--) self::$area[$r][] = $p;
        }
        foreach (self::$area as $k=>$a) usort (self::$area[$k], function ($a,$b) { $d = $a->norm()-$b->norm(); return ($d < 0) ? -1 : (($d > 0) ? 1 : 0); });
    }
}
Border::init(2.5,6,8);

class Map {
    static
        $waypoint_lookup, // retrieve waypoint from a position
        $dist_to_goal;    // upper bound of distance to closest goal for each track point
                          // also used to check if a point is on track

/*      
    const NONE  = 0;  // terrain types
    const GOAL  = 1;
    const START = 2;
    const TRACK = 3;
*/      
    static function solve ($filename)
    {
        // read map definition
        $img = imagecreatefrompng ($filename);// or die ("could not read $filename");
        $img_dx = imagesx ($img);
        $img_dy = imagesy ($img);

        // extract track pixels
        for ($x = 0 ; $x != $img_dx ; $x++)
        for ($y = 0 ; $y != $img_dy ; $y++)
        {
            $type = 0 /*Map::NONE*/;
            $color = imagecolorat ($img, $x, $y);
            if      ($color  ==        0) { $goals [] = new Point ($x, $y); $type = 1 /*Map::GOAL*/;  }
            else if ($color  == 0xFFFF00) { $starts[] = new Point ($x, $y); $type = 2 /*Map::START*/; }
            else
            {
                $r = ($color >> 16) & 0xFF;
                $g = ($color >>  8) & 0xFF;
                $b =  $color        & 0xFF;
                if ($r == $g && $r == $b && $r >= 30 && $r <= 220) $type = 3 /*Map::TRACK*/;
            }
            if ($type /* != Map::NONE */)
            {
                $track[$x][$y] = 1; // mark all track points
                if ($type != 3 /*Map::TRACK*/) $waypoints[] = new Waypoint ($x, $y, $type); // add start and goal positions as waypoints
            }
        }

        // compute possible acceleration values
        for ($x = -ACCEL_MAX ; $x <= ACCEL_MAX ; $x++)
        for ($y = -ACCEL_MAX ; $y <= ACCEL_MAX ; $y++)
            if (abs ($x) + abs ($y) <= ACCEL_MAX) $acceleration[] = new Point ($x, $y);

        // compute goal distance
        $neighbours = array (new Point (-1, 0), new Point (1, 0), new Point (0, -1), new Point (0, 1)); 
        foreach ($goals as $goal)
        {
            $border[] = new Point ($goal->x, $goal->y);
            Map::$dist_to_goal[$goal->x][$goal->y] = 0;
        }
        while (count ($border))
        {
            $pos = array_shift ($border);
            $dist = Map::$dist_to_goal[$pos->x][$pos->y] + 1;
            foreach ($neighbours as $n)
            {
                $nx = $pos->x + $n->x;
                $ny = $pos->y + $n->y;
                if (!isset ($track[$nx][$ny])) continue;
                if (isset (Map::$dist_to_goal[$nx][$ny])) continue;
                Map::$dist_to_goal[$nx][$ny] = $dist;
                $border[] = new Point ($nx, $ny);
            }
        }

        // compute wall distance vector field
        foreach (self::$dist_to_goal as $x=>$col)
        foreach ($col as $y=>$c)
        {
            $num = 0;
            $ix = $iy = 0;
            foreach (Border::$circle as $d=>$points)
            {
                foreach ($points as $p)
                {
                    if (!isset (Map::$dist_to_goal[$x+$p->x][$y+$p->y]))
                    {
                        $ix += $p->ix;
                        $iy += $p->iy;
                        $num++;
                    }
                }
                if ($num != 0)
                {
                    $ix /= $num;
                    $iy /= $num;
                    break;
                }
            }
            $wall_vector[$x][$y] = new Point ($ix, $iy);
        }

        // compute local curvature and deduce desired waypoint density
        foreach (self::$dist_to_goal as $x=>$col)
        foreach ($col as $y=>$c)
        {
            $o = $wall_vector[$x][$y];
            $oo = $o->normalized();
            $curvature = 0;
            $num = 0;
            foreach (Border::$area[1] as $n)
            {
                @$p = $wall_vector[$x+$n->x][$y+$n->y];
                if (!$p) continue;
                $pp = $p->normalized();
                $nn = $n->normalized();
                $curvature -= ($oo->x*$pp->y-$oo->y*$pp->x) * ($oo->x*$nn->y-$oo->y*$nn->x);
                $num++;
            }
            $waypoint_density[$x][$y] = (12*$curvature/$num + 1) * $o->norm(); // a wierd mix of curvature and wall distance
        }

        // compute track waypoints
        $rmin = 1;
        $rmax = 6;
        $c_max = 0;
        foreach ($waypoint_density as $x=>$col)
        foreach ($col as $y=>$c)
            $c_max = max ($c_max, $c);
        $ra = ($rmin-$rmax)/$c_max;
        foreach ($waypoint_density as $x=>$col)
        foreach ($col as $y=>$c)
            $placement[($rmax + $ra * max ($c, 0))*1e5][] = new Point ($x, $y);
        ksort($placement);
//var_dump($placement);exit(0);
        foreach ($placement as $r=>$points)
        foreach ($points as $p)
        {
            $x = $p->x;
            $y = $p->y;
            if (!isset ($waypoint_density[$x][$y])) continue;
            $waypoints[] = new Waypoint ($x, $y, 3 /*Map::TRACK*/);
            unset ($waypoint_density[$x][$y]);
            $out=0;
            foreach (Border::$area[$r/1e5] as $delta)
            {
                $nx = $x+$delta->x;
                $ny = $y+$delta->y;
                if (on_track ($x, $y, $nx, $ny)) unset ($waypoint_density[$nx][$ny]);
                else if (++$out == 2) break;
            }
        }

        // unleash the mighty A*
//$begining=microtime(true);
        foreach ($starts as $start)
        {
            $n = new Node ($start->x, $start->y);
            $border[$n->moves+$n->dist][] = $n;
            $open[$start->x." ".$start->y." 0 0"] = $n;
        }
        ksort ($border);
        while (count ($border))
        {
            // get one of the most prioritary nodes
            reset ($border);
            $p_list = key ($border);
            $node = array_shift ($border[$p_list]);
            if (empty ($border[$p_list])) unset ($border[$p_list]);

            $px = $node->x;
            $py = $node->y;
            $vx = $node->speedx;
            $vy = $node->speedy;
            $current = Map::$waypoint_lookup[$px][$py];

            // move node from open to closed list
            $signature = "$px $py $vx $vy";
            unset ($open[$signature]);
            $closed[$signature] = 1;

            // try all possible accelerations
            foreach ($acceleration as $a)
            {
                $nvx = $vx + $a->x;
                $nvy = $vy + $a->y;
                $npx = $px + $nvx;
                $npy = $py + $nvy;

                // select waypoints within reach
                @$waypoint = Map::$waypoint_lookup[$npx][$npy];
                if (!$waypoint) continue;

                // skip already know nodes
                $signature = "$npx $npy $nvx $nvy";
                if (@$closed[$signature] || @$open[$signature]) continue;

                // check track geometry
                if (!$waypoint->can_reach ($current)) continue;

                // insert new node into priority list
                $nn = new Node ($npx, $npy, $nvx, $nvy, $node);
                $p = $nn->moves+$nn->dist;
                $resort = !isset($border[$p]);
                $border[$p][] = $nn;
                if ($resort) ksort ($border);
                $open[$signature] = $nn;

                // check termination
                if ($waypoint->type == 1 /*Map::GOAL*/)
                {
                    for ($path=array() ; $nn != null ; $nn = $nn->parent) array_unshift ($path, $nn);
                    $ox = $path[0]->x;
                    $oy = $path[0]->y;
                    for ($i = 1 ; $i != count($path) ; $i++)
                    {
                        $ex = $path[$i]->x;
                        $ey = $path[$i]->y;
                        imageline ($img, $ox, $oy, $ex, $ey, 0xFFFFFF);
                        $ox = $ex; $oy = $ey;
                    }
                    foreach ($path as $p) imagefilledellipse ($img, $p->x, $p->y, 2, 2, 0xFF);
                    imagepng ($img, "_".$filename);
//echo (count($path)-1)." moves, ".count($waypoints)." waypoints, ".count($closed)."+".count($open)." nodes, ".(round((microtime(true)-$begining)*100)/100)."s, ".round(memory_get_usage(true)/1024)."K";
                    return;
                }
            }
        }
    }
}

ini_set("memory_limit","2G"); // just in case...
Map::solve ($argv[1]);
?>

Results

The pics are produced by a richer version that outputs the exact same solution with some statistics at the bottom (and draws the path with antialiasing).

The city map is a pretty good example of why position-based algorithms are bound to find subpar results in many cases: shorter does not always mean faster.

city track obstacles nightmare (672 moves if you don't want to zoom)

A*

To my surprise, A* performs rather well on the position-speed space. Better than BFS, at any rate.

I had to sweat quite a bit to produce a working heuristic distance estimate, though.

I had to optimize it somewhat too, since the number of possible states is huge (a few millions) compared with a position-only variant, which again requires more code.

The lower bound chosen for number of moves necessary to reach a goal from a given position is simply the time needed to cover the distance to the nearest goal in a straight line with zero initial speed.

Of course, the straight line path will usually lead directly into a wall, but this is about the same problem as using the euclidian straight distance for space-only A* searchs.
Just like the euclidian distance for space-only variants, the main advantage of this metric, besides being acceptable to use the most efficient A* variant (with closed nodes), is to require very little topological analysis of the track.

Given a maximal acceleration A, the number n of moves needed to cover a distance d is the smallest integer that satisfies the relation:

d <= A n(n+1)/2

Solving this for n gives an estimate of the remaining distance.

To compute this, a map of distance to nearest goal is built, using a flood-fill algorithm seeded with goal positions.
It has the pleasant side effect of eliminating track points from where no goal can be reached (as it happens in some areas of the nightmare track).
The number of moves is computed as a floating point value, so that nodes nearer to the goal can be further discriminated.

Waypoints

As in my previous attempt, the idea is to reduce the number of track points to an as small as possible subsample of waypoints. The trick is to try and pick the most "useful" positions on the track.

The heuristic starts with a regular repartition over the whole track, but increases the density in two types of areas :

  1. the rim of the track, i.e. lanes running 1 or 2 pixels away from walls
  2. zones of high curvature, i.e. the inner rim of sharp bends.

Here is an example.
High density areas are in red, low density in green. Blue pixels are the selected waypoints.
Notice the clusters of waypoints on sharp bends (among a lot of useless blotches on slanted curves, due to insufficient filtering).

waypoints density and placement

To compute lanes positions, the entire track is scanned for distance to nearest wall. The result is a field of vectors pointing toward the nearest track border.
This vector field is then processed to produce a rough estimate of local curvature.
Finally, curvature and distance to wall are combined to produce a desired local density, and a rather clunky algorithm tries to sprinkle waypoints over the track accordingly.

A notable improvement over the previous strategy is that narrow lanes will (apparently) always get enough waypoints to drive through, which allows to navigate the nightmare map.

As always, it is a matter of finding a sweet spot between computation time and efficiency.
A 50% decrease in density will divide computation time by more than 4, but with coarser results (48 moves instead of 44 on city, 720 instead of 670 on nightmare).

Golfing

I still think golfing does harm creativity in this specific case: removing antialiasing from the output is enough to gain 30 points, and requires a lot less effort than going from 47 to 44 moves on the city map.
Even going from 720 to 670 moves on nightmare would gain a mere 500 points, though I very doubt a position-only A* would be able to go anywhere near that.

Just for the fun of it, I decided to write my own PHP compressor anyway.

As it appears, renaming identifiers efficiently in PHP is not an easy task. In fact, I do not think it is even possible to do it in the general case. Even with a full semantic analysis, the possibility to use strings or indirect variables to designate objects would require the knowledge of every function semantics.
However, since the very convenient built-in parser allows to jump to semantic analysis right away, I managed to produce something that seems to work on a subset of PHP sufficient to write "golfable" code (stay away from $$ and don't use indirect function calls or other string access to objects).
No practical use to speak of and nothing to do with the original problem, but a lot of fun to code nevertheless.

I could have butchered the code further to gain an extra 500 characters or so, but since the names of PHP graphic library are unfortunately quite long, it's a kind of uphill struggle.

Further developments

The waypoint selection code is an horrendous mess, tuned by trial and error. I suspect doing more maths (using proper gradient and curl operators) would greatly enhance the process.

I am curious to see if a better distance heuristic can be found. I did try to take speed into account in a couple of ways, but it either broke the A* or produced slower results.

It might be possible to recode all this in a faster language like C++, but the PHP version relies heavily on garbage collection to keep memory consumption reasonable. Cleaning up closed nodes in C++ would require quite a bit of work and a sizeable amount of extra code.

Would the scoring have been based on performances, I would have eagerly tried to improve the algorithms. But since the golfing criterion is so overwhelming, there is no real point, or is it?

user16991

Posted 2014-12-06T07:27:26.907

Reputation:

Impressive algorithm work, and excellent description of your entry +1. – Logic Knight – 2015-01-14T06:16:27.363

Hopefully the offered bonus will answer your last question. Let's see how fast these cars can go! – Logic Knight – 2015-01-15T06:23:20.930

Hehe I have a slightly faster version, but it's so ugly I would rather not publish it unless someone beats the current one :) – None – 2015-01-15T12:20:59.120

You are in the running for the bounty but you need to show the valid path list to claim it (see edited question). – Logic Knight – 2015-01-19T14:28:53.060

Bah, my piece of junk of a code is not worth any bounty. I would rather hope for a better solution to emerge. – None – 2015-01-19T16:07:19.907

@kuroineko: C'mon, post your path! It's not your code that is judged it's the result. – nimi – 2015-01-20T17:07:23.417

I have awarded you the bounty. Your performance on the Gauntlet is excellent! – Logic Knight – 2015-01-21T05:44:54.080

Thanks guv'. Too bad no serious contender showed up, though – None – 2015-01-23T01:51:36.060

2

ThirdRacer Java (1224 + 93*10 = 2154)

Similar to SecondRacer. But switching focus from speed to code size (but still using Java). Optimizing the acceleration is much simplified now, sadly resulting in a slower car.

Performance

Better than SecondRacer.

Path Style

Like SecondRacer.

Code Style

I entered into a heavy fighting mode.

golfed -> WARNING: it does in-place replacement of the original file!

import javax.imageio.*;class A{class B extends java.util.Vector<C>{};class
C{int D,E;}C F(int D,int E){G=new C();G.D=D;G.E=E;return G;}static java.awt.image.BufferedImage
H;int I=H.getWidth(),J=H.getHeight(),K[][]=new int[I][J],L,M,N,O,P=~0xffff00,Q,D,E,R,S,T,U,V=255,W,X,Y;C
Z,G;public static void main(String[]a)throws Exception{java.io.File b=new
java.io.File(a[0]);H=ImageIO.read(b);new A().c();ImageIO.write(H,"PNG",b);}void
c(){B d=new B();for(L=0;L<I;L++)for(M=0;M<J;M++)if(e(L,M)!=1||!d.add(F(L,M)))K[L][M]=-1>>>1;while(M!=3)for(Z=d.remove(N=0),D=Z.D,E=Z.E;N<9;N++)if((M=e(T=D+N/3-1,U=E+N%3-1))>0&&K[T][U]>(L=K[D][E]+(T==D||U==E?10:14))&&d.add(F(T,U)))K[T][U]=L;for(D=G.D,E=G.E,R=D,S=E;M!=4;){H.createGraphics().drawLine(R,S,D,E);H.setRGB(R,S,P);N=0;T=2-M%2;U=0;for(L=0;L<Q;L++,N+=T)if((N+T)*(N+T)/30.0>Q-L+7||N-O>15-T){H.setRGB(R+L*(M/3-1),S+L*(M%3-1),P);U=L;O=N;N=0;}O=T*(U-Q);R=D;S=E;M=4;double
f=0,g;for(N=0;N<9;N++)for(L=1;e(T=R+L*(N/3-1),U=S+L*(N%3-1))>0;L++)if(f>(g=K[T][U]-K[R][S]+5*java.lang.Math.sqrt((R-T)*(R-T)+(S-U)*(S-U)))){f=g;D=T;E=U;M=N;Q=L;}}H.setRGB(R,S,P);}int
e(int D,int E){return D<0||D>=I||E<0||E>=J?0:(W=H.getRGB(D,E))==~V?1:W==V<<24?3:30<=(X=W>>16&V)&&X<=220&&X==(Y=W>>8&V)&&Y==(V&W)?2:0;}}

City S+93

City S+93

Bob Genom

Posted 2014-12-06T07:27:26.907

Reputation: 846

Well done! How long does it take to run the program? – nimi – 2015-01-16T20:02:17.340

City: 2146ms and Gauntlet: 9643ms. But more than half of the time is spend in ImageIO.write(..) writing the image to disk. It's so fast because it's NOT exploring the position+speed space. – Bob Genom – 2015-01-17T11:04:55.757

1

Star crossed racer path on the nightmare map

(as per popular request)

(code not updated since the modifications are trivial and the performance-only challenge is not golfed)

Sorry to post yet another entry, but I'm hitting the 30.000 characters limit on the previous one.
Just say the word and I'll delete this one.

  1: 112 154 -> 127 154
  2: 127 154 -> 142 154
  3: 142 154 -> 151 161
  4: 151 161 -> 149 171
  5: 149 171 -> 143 190
  6: 143 190 -> 131 208
  7: 131 208 -> 125 219
  8: 125 219 -> 132 230
  9: 132 230 -> 147 243
 10: 147 243 -> 169 249
 11: 169 249 -> 185 248
 12: 185 248 -> 190 251
 13: 190 251 -> 190 263
 14: 190 263 -> 194 282
 15: 194 282 -> 201 289
 16: 201 289 -> 219 299
 17: 219 299 -> 240 297
 18: 240 297 -> 256 289
 19: 256 289 -> 271 267
 20: 271 267 -> 283 241
 21: 283 241 -> 297 228
 22: 297 228 -> 315 226
 23: 315 226 -> 343 229
 24: 343 229 -> 370 246
 25: 370 246 -> 393 263
 26: 393 263 -> 415 270
 27: 415 270 -> 435 267
 28: 435 267 -> 454 251
 29: 454 251 -> 464 240
 30: 464 240 -> 468 238
 31: 468 238 -> 472 247
 32: 472 247 -> 475 270
 33: 475 270 -> 481 302
 34: 481 302 -> 489 323
 35: 489 323 -> 489 343
 36: 489 343 -> 476 365
 37: 476 365 -> 455 380
 38: 455 380 -> 437 389
 39: 437 389 -> 432 398
 40: 432 398 -> 437 405
 41: 437 405 -> 450 411
 42: 450 411 -> 462 430
 43: 462 430 -> 465 454
 44: 465 454 -> 457 482
 45: 457 482 -> 453 503
 46: 453 503 -> 460 523
 47: 460 523 -> 469 530
 48: 469 530 -> 485 530
 49: 485 530 -> 505 526
 50: 505 526 -> 514 522
 51: 514 522 -> 523 533
 52: 523 533 -> 526 552
 53: 526 552 -> 527 572
 54: 527 572 -> 531 581
 55: 531 581 -> 535 577
 56: 535 577 -> 539 559
 57: 539 559 -> 542 527
 58: 542 527 -> 544 481
 59: 544 481 -> 550 425
 60: 550 425 -> 558 356
 61: 558 356 -> 565 296
 62: 565 296 -> 572 250
 63: 572 250 -> 575 213
 64: 575 213 -> 575 188
 65: 575 188 -> 565 168
 66: 565 168 -> 567 147
 67: 567 147 -> 569 141
 68: 569 141 -> 574 144
 69: 574 144 -> 582 158
 70: 582 158 -> 587 160
 71: 587 160 -> 592 148
 72: 592 148 -> 593 139
 73: 593 139 -> 597 141
 74: 597 141 -> 605 151
 75: 605 151 -> 616 165
 76: 616 165 -> 616 177
 77: 616 177 -> 609 181
 78: 609 181 -> 599 174
 79: 599 174 -> 592 168
 80: 592 168 -> 591 171
 81: 591 171 -> 589 188
 82: 589 188 -> 591 216
 83: 591 216 -> 595 257
 84: 595 257 -> 599 312
 85: 599 312 -> 605 367
 86: 605 367 -> 611 408
 87: 611 408 -> 614 438
 88: 614 438 -> 609 461
 89: 609 461 -> 597 477
 90: 597 477 -> 594 499
 91: 594 499 -> 604 520
 92: 604 520 -> 605 536
 93: 605 536 -> 598 556
 94: 598 556 -> 598 569
 95: 598 569 -> 610 580
 96: 610 580 -> 622 581
 97: 622 581 -> 629 582
 98: 629 582 -> 636 568
 99: 636 568 -> 642 541
100: 642 541 -> 645 526
101: 645 526 -> 645 517
102: 645 517 -> 634 505
103: 634 505 -> 636 493
104: 636 493 -> 639 467
105: 639 467 -> 641 427
106: 641 427 -> 644 373
107: 644 373 -> 648 309
108: 648 309 -> 651 258
109: 651 258 -> 652 218
110: 652 218 -> 652 190
111: 652 190 -> 647 167
112: 647 167 -> 645 147
113: 645 147 -> 645 138
114: 645 138 -> 655 134
115: 655 134 -> 670 137
116: 670 137 -> 675 142
117: 675 142 -> 676 156
118: 676 156 -> 679 168
119: 679 168 -> 680 178
120: 680 178 -> 667 188
121: 667 188 -> 661 195
122: 661 195 -> 663 208
123: 663 208 -> 667 233
124: 667 233 -> 671 271
125: 671 271 -> 676 322
126: 676 322 -> 681 386
127: 681 386 -> 687 445
128: 687 445 -> 693 492
129: 693 492 -> 695 530
130: 695 530 -> 698 554
131: 698 554 -> 701 565
132: 701 565 -> 704 564
133: 704 564 -> 707 548
134: 707 548 -> 709 518
135: 709 518 -> 710 474
136: 710 474 -> 716 420
137: 716 420 -> 720 355
138: 720 355 -> 724 305
139: 724 305 -> 724 266
140: 724 266 -> 726 239
141: 726 239 -> 727 225
142: 727 225 -> 729 224
143: 729 224 -> 732 235
144: 732 235 -> 734 260
145: 734 260 -> 734 296
146: 734 296 -> 734 347
147: 734 347 -> 734 413
148: 734 413 -> 734 479
149: 734 479 -> 734 533
150: 734 533 -> 735 573
151: 735 573 -> 735 599
152: 735 599 -> 732 616
153: 732 616 -> 729 618
154: 729 618 -> 713 618
155: 713 618 -> 683 618
156: 683 618 -> 638 618
157: 638 618 -> 578 618
158: 578 618 -> 503 618
159: 503 618 -> 413 618
160: 413 618 -> 320 618
161: 320 618 -> 242 618
162: 242 618 -> 179 618
163: 179 618 -> 131 618
164: 131 618 ->  98 618
165:  98 618 ->  80 618
166:  80 618 ->  72 617
167:  72 617 ->  69 606
168:  69 606 ->  69 585
169:  69 585 ->  69 549
170:  69 549 ->  69 498
171:  69 498 ->  69 432
172:  69 432 ->  69 351
173:  69 351 ->  69 276
174:  69 276 ->  69 216
175:  69 216 ->  69 171
176:  69 171 ->  69 141
177:  69 141 ->  69 126
178:  69 126 ->  75 118
179:  75 118 ->  87 118
180:  87 118 -> 114 118
181: 114 118 -> 156 118
182: 156 118 -> 213 118
183: 213 118 -> 285 118
184: 285 118 -> 372 118
185: 372 118 -> 474 118
186: 474 118 -> 591 118
187: 591 118 -> 701 120
188: 701 120 -> 800 120
189: 800 120 -> 884 120
190: 884 120 -> 953 120
191: 953 120 -> 1007 120
192: 1007 120 -> 1049 120
193: 1049 120 -> 1076 120
194: 1076 120 -> 1089 120
195: 1089 120 -> 1092 123
196: 1092 123 -> 1087 132
197: 1087 132 -> 1073 145
198: 1073 145 -> 1046 160
199: 1046 160 -> 1015 164
200: 1015 164 -> 986 156
201: 986 156 -> 964 150
202: 964 150 -> 954 147
203: 954 147 -> 951 151
204: 951 151 -> 959 156
205: 959 156 -> 981 162
206: 981 162 -> 996 169
207: 996 169 -> 1002 182
208: 1002 182 -> 997 194
209: 997 194 -> 986 208
210: 986 208 -> 988 222
211: 988 222 -> 995 226
212: 995 226 -> 1013 226
213: 1013 226 -> 1044 224
214: 1044 224 -> 1079 229
215: 1079 229 -> 1103 238
216: 1103 238 -> 1119 245
217: 1119 245 -> 1133 243
218: 1133 243 -> 1147 256
219: 1147 256 -> 1153 270
220: 1153 270 -> 1160 270
221: 1160 270 -> 1162 260
222: 1162 260 -> 1165 237
223: 1165 237 -> 1182 213
224: 1182 213 -> 1210 185
225: 1210 185 -> 1231 157
226: 1231 157 -> 1245 135
227: 1245 135 -> 1257 123
228: 1257 123 -> 1261 118
229: 1261 118 -> 1263 124
230: 1263 124 -> 1263 143
231: 1263 143 -> 1263 176
232: 1263 176 -> 1263 224
233: 1263 224 -> 1263 287
234: 1263 287 -> 1263 365
235: 1263 365 -> 1263 437
236: 1263 437 -> 1263 494
237: 1263 494 -> 1263 536
238: 1263 536 -> 1263 563
239: 1263 563 -> 1263 578
240: 1263 578 -> 1258 583
241: 1258 583 -> 1243 583
242: 1243 583 -> 1213 583
243: 1213 583 -> 1180 580
244: 1180 580 -> 1146 568
245: 1146 568 -> 1125 558
246: 1125 558 -> 1117 546
247: 1117 546 -> 1115 539
248: 1115 539 -> 1107 538
249: 1107 538 -> 1098 550
250: 1098 550 -> 1103 561
251: 1103 561 -> 1114 567
252: 1114 567 -> 1113 575
253: 1113 575 -> 1099 581
254: 1099 581 -> 1078 582
255: 1078 582 -> 1067 579
256: 1067 579 -> 1059 570
257: 1059 570 -> 1061 560
258: 1061 560 -> 1070 556
259: 1070 556 -> 1074 553
260: 1074 553 -> 1069 544
261: 1069 544 -> 1058 542
262: 1058 542 -> 1045 530
263: 1045 530 -> 1017 518
264: 1017 518 -> 990 509
265: 990 509 -> 972 501
266: 972 501 -> 955 500
267: 955 500 -> 938 514
268: 938 514 -> 914 528
269: 914 528 -> 902 543
270: 902 543 -> 895 562
271: 895 562 -> 893 572
272: 893 572 -> 880 581
273: 880 581 -> 869 579
274: 869 579 -> 858 571
275: 858 571 -> 844 567
276: 844 567 -> 834 558
277: 834 558 -> 830 553
278: 830 553 -> 832 540
279: 832 540 -> 829 529
280: 829 529 -> 821 522
281: 821 522 -> 819 517
282: 819 517 -> 831 512
283: 831 512 -> 838 506
284: 838 506 -> 843 488
285: 843 488 -> 843 473
286: 843 473 -> 844 469
287: 844 469 -> 856 469
288: 856 469 -> 883 469
289: 883 469 -> 906 458
290: 906 458 -> 918 449
291: 918 449 -> 924 433
292: 924 433 -> 920 418
293: 920 418 -> 904 406
294: 904 406 -> 883 404
295: 883 404 -> 859 402
296: 859 402 -> 844 394
297: 844 394 -> 843 385
298: 843 385 -> 841 366
299: 841 366 -> 838 361
300: 838 361 -> 828 363
301: 828 363 -> 813 356
302: 813 356 -> 807 343
303: 807 343 -> 805 321
304: 805 321 -> 810 298
305: 810 298 -> 813 285
306: 813 285 -> 821 282
307: 821 282 -> 842 280
308: 842 280 -> 868 278
309: 868 278 -> 887 280
310: 887 280 -> 898 288
311: 898 288 -> 898 300
312: 898 300 -> 895 314
313: 895 314 -> 901 324
314: 901 324 -> 909 324
315: 909 324 -> 917 318
316: 917 318 -> 921 311
317: 921 311 -> 930 314
318: 930 314 -> 947 322
319: 947 322 -> 956 329
320: 956 329 -> 962 339
321: 962 339 -> 970 337
322: 970 337 -> 973 338
323: 973 338 -> 978 334
324: 978 334 -> 992 326
325: 992 326 -> 1000 327
326: 1000 327 -> 1008 335
327: 1008 335 -> 1015 351
328: 1015 351 -> 1021 373
329: 1021 373 -> 1022 390
330: 1022 390 -> 1013 404
331: 1013 404 -> 1006 417
332: 1006 417 -> 1012 430
333: 1012 430 -> 1023 436
334: 1023 436 -> 1029 434
335: 1029 434 -> 1049 432
336: 1049 432 -> 1063 426
337: 1063 426 -> 1079 425
338: 1079 425 -> 1093 418
339: 1093 418 -> 1113 417
340: 1113 417 -> 1128 414
341: 1128 414 -> 1139 421
342: 1139 421 -> 1154 426
343: 1154 426 -> 1158 430
344: 1158 430 -> 1149 436
345: 1149 436 -> 1130 438
346: 1130 438 -> 1108 442
347: 1108 442 -> 1096 447
348: 1096 447 -> 1087 441
349: 1087 441 -> 1079 443
350: 1079 443 -> 1072 446
351: 1072 446 -> 1060 454
352: 1060 454 -> 1052 461
353: 1052 461 -> 1034 463
354: 1034 463 -> 1016 463
355: 1016 463 -> 1010 464
356: 1010 464 -> 1011 472
357: 1011 472 -> 1012 479
358: 1012 479 -> 1025 484
359: 1025 484 -> 1048 488
360: 1048 488 -> 1083 491
361: 1083 491 -> 1119 505
362: 1119 505 -> 1154 520
363: 1154 520 -> 1183 530
364: 1183 530 -> 1201 537
365: 1201 537 -> 1209 539
366: 1209 539 -> 1209 535
367: 1209 535 -> 1209 517
368: 1209 517 -> 1209 484
369: 1209 484 -> 1210 437
370: 1210 437 -> 1210 392
371: 1210 392 -> 1210 362
372: 1210 362 -> 1210 347
373: 1210 347 -> 1203 340
374: 1203 340 -> 1184 333
375: 1184 333 -> 1156 320
376: 1156 320 -> 1116 306
377: 1116 306 -> 1069 285
378: 1069 285 -> 1023 265
379: 1023 265 -> 985 249
380: 985 249 -> 955 235
381: 955 235 -> 933 227
382: 933 227 -> 923 221
383: 923 221 -> 923 211
384: 923 211 -> 917 195
385: 917 195 -> 901 176
386: 901 176 -> 881 159
387: 881 159 -> 848 144
388: 848 144 -> 815 144
389: 815 144 -> 788 153
390: 788 153 -> 769 169
391: 769 169 -> 764 185
392: 764 185 -> 766 209
393: 766 209 -> 767 247
394: 767 247 -> 769 299
395: 769 299 -> 769 362
396: 769 362 -> 769 440
397: 769 440 -> 769 503
398: 769 503 -> 769 551
399: 769 551 -> 769 584
400: 769 584 -> 769 605
401: 769 605 -> 770 613
402: 770 613 -> 780 616
403: 780 616 -> 801 616
404: 801 616 -> 837 616
405: 837 616 -> 888 616
406: 888 616 -> 954 616
407: 954 616 -> 1035 616
408: 1035 616 -> 1113 616
409: 1113 616 -> 1176 616
410: 1176 616 -> 1224 616
411: 1224 616 -> 1257 616
412: 1257 616 -> 1278 616
413: 1278 616 -> 1294 607
414: 1294 607 -> 1295 598
415: 1295 598 -> 1295 577
416: 1295 577 -> 1295 541
417: 1295 541 -> 1295 490
418: 1295 490 -> 1295 424
419: 1295 424 -> 1295 343
420: 1295 343 -> 1295 265
421: 1295 265 -> 1295 202
422: 1295 202 -> 1295 154
423: 1295 154 -> 1295 121
424: 1295 121 -> 1295 100
425: 1295 100 -> 1294  92
426: 1294  92 -> 1283  89
427: 1283  89 -> 1262  89
428: 1262  89 -> 1226  89
429: 1226  89 -> 1175  89
430: 1175  89 -> 1109  89
431: 1109  89 -> 1028  89
432: 1028  89 -> 938  89
433: 938  89 -> 860  89
434: 860  89 -> 797  89
435: 797  89 -> 749  89
436: 749  89 -> 716  89
437: 716  89 -> 698  89
438: 698  89 -> 690  94
439: 690  94 -> 682 102
440: 682 102 -> 673 100
441: 673 100 -> 661  89
442: 661  89 -> 646  89
443: 646  89 -> 616  89
444: 616  89 -> 571  89
445: 571  89 -> 517  89
446: 517  89 -> 478  89
447: 478  89 -> 454  89
448: 454  89 -> 442  92
449: 442  92 -> 432 102
450: 432 102 -> 423 100
451: 423 100 -> 411  89
452: 411  89 -> 396  89
453: 396  89 -> 381  92
454: 381  92 -> 371 102
455: 371 102 -> 362 100
456: 362 100 -> 349  89
457: 349  89 -> 334  89
458: 334  89 -> 309  91
459: 309  91 -> 298  92
460: 298  92 -> 288 102
461: 288 102 -> 279 100
462: 279 100 -> 267  89
463: 267  89 -> 252  89
464: 252  89 -> 222  89
465: 222  89 -> 177  89
466: 177  89 -> 123  89
467: 123  89 ->  84  89
468:  84  89 ->  60  89
469:  60  89 ->  48  89
470:  48  89 ->  42  97
471:  42  97 ->  42 112
472:  42 112 ->  42 142
473:  42 142 ->  42 187
474:  42 187 ->  42 247
475:  42 247 ->  42 322
476:  42 322 ->  42 409
477:  42 409 ->  42 484
478:  42 484 ->  42 544
479:  42 544 ->  42 589
480:  42 589 ->  42 619
481:  42 619 ->  42 634
482:  42 634 ->  47 640
483:  47 640 ->  59 640
484:  59 640 ->  86 640
485:  86 640 -> 128 640
486: 128 640 -> 185 640
487: 185 640 -> 257 640
488: 257 640 -> 344 640
489: 344 640 -> 446 640
490: 446 640 -> 563 640
491: 563 640 -> 690 640
492: 690 640 -> 816 639
493: 816 639 -> 930 639
494: 930 639 -> 1029 639
495: 1029 639 -> 1113 639
496: 1113 639 -> 1182 639
497: 1182 639 -> 1236 639
498: 1236 639 -> 1275 639
499: 1275 639 -> 1300 639
500: 1300 639 -> 1316 633
501: 1316 633 -> 1320 630
502: 1320 630 -> 1322 615
503: 1322 615 -> 1322 588
504: 1322 588 -> 1322 546
505: 1322 546 -> 1322 489
506: 1322 489 -> 1322 417
507: 1322 417 -> 1322 330
508: 1322 330 -> 1322 252
509: 1322 252 -> 1322 186
510: 1322 186 -> 1322 135
511: 1322 135 -> 1322  99
512: 1322  99 -> 1322  78
513: 1322  78 -> 1320  68
514: 1320  68 -> 1310  65
515: 1310  65 -> 1289  65
516: 1289  65 -> 1253  65
517: 1253  65 -> 1208  65
518: 1208  65 -> 1178  65
519: 1178  65 -> 1163  65
520: 1163  65 -> 1155  71
521: 1155  71 -> 1149  76
522: 1149  76 -> 1135  74
523: 1135  74 -> 1111  74
524: 1111  74 -> 1102  74
525: 1102  74 -> 1091  65
526: 1091  65 -> 1076  65
527: 1076  65 -> 1046  65
528: 1046  65 -> 1013  65
529: 1013  65 -> 992  65
530: 992  65 -> 986  65
531: 986  65 -> 975  56
532: 975  56 -> 960  56
533: 960  56 -> 930  56
534: 930  56 -> 899  58
535: 899  58 -> 878  58
536: 878  58 -> 870  59
537: 870  59 -> 864  65
538: 864  65 -> 849  65
539: 849  65 -> 819  65
540: 819  65 -> 774  65
541: 774  65 -> 714  65
542: 714  65 -> 651  65
543: 651  65 -> 603  65
544: 603  65 -> 570  65
545: 570  65 -> 552  65
546: 552  65 -> 546  65
547: 546  65 -> 535  56
548: 535  56 -> 520  56
549: 520  56 -> 492  58
550: 492  58 -> 478  59
551: 478  59 -> 472  65
552: 472  65 -> 457  65
553: 457  65 -> 427  65
554: 427  65 -> 382  65
555: 382  65 -> 322  65
556: 322  65 -> 265  65
557: 265  65 -> 223  65
558: 223  65 -> 193  65
559: 193  65 -> 178  65
560: 178  65 -> 170  71
561: 170  71 -> 164  76
562: 164  76 -> 156  74
563: 156  74 -> 145  65
564: 145  65 -> 130  65
565: 130  65 -> 109  65
566: 109  65 -> 103  65
567: 103  65 ->  92  56
568:  92  56 ->  77  56
569:  77  56 ->  65  59
570:  65  59 ->  57  68
571:  57  68 ->  46  67
572:  46  67 ->  29  65
573:  29  65 ->  20  68
574:  20  68 ->  17  80
575:  17  80 ->  17 104
576:  17 104 ->  17 143
577:  17 143 ->  17 197
578:  17 197 ->  17 266
579:  17 266 ->  17 350
580:  17 350 ->  19 435
581:  19 435 ->  19 507
582:  19 507 ->  19 564
583:  19 564 ->  19 606
584:  19 606 ->  19 633
585:  19 633 ->  19 648
586:  19 648 ->  33 662
587:  33 662 ->  48 664
588:  48 664 ->  76 664
589:  76 664 -> 118 664
590: 118 664 -> 175 664
591: 175 664 -> 245 664
592: 245 664 -> 328 664
593: 328 664 -> 423 663
594: 423 663 -> 532 661
595: 532 661 -> 654 661
596: 654 661 -> 784 662
597: 784 662 -> 900 662
598: 900 662 -> 1002 662
599: 1002 662 -> 1089 662
600: 1089 662 -> 1166 662
601: 1166 662 -> 1231 664
602: 1231 664 -> 1283 664
603: 1283 664 -> 1320 664
604: 1320 664 -> 1344 662
605: 1344 662 -> 1355 662
606: 1355 662 -> 1359 654
607: 1359 654 -> 1372 640
608: 1372 640 -> 1377 630
609: 1377 630 -> 1376 613
610: 1376 613 -> 1376 586
611: 1376 586 -> 1376 550
612: 1376 550 -> 1374 527
613: 1374 527 -> 1374 517
614: 1374 517 -> 1381 508
615: 1381 508 -> 1381 494
616: 1381 494 -> 1370 477
617: 1370 477 -> 1372 459
618: 1372 459 -> 1370 432
619: 1370 432 -> 1367 418
620: 1367 418 -> 1354 401
621: 1354 401 -> 1355 384
622: 1355 384 -> 1351 361
623: 1351 361 -> 1343 330
624: 1343 330 -> 1344 295
625: 1344 295 -> 1346 271
626: 1346 271 -> 1347 256
627: 1347 256 -> 1336 240
628: 1336 240 -> 1336 224
629: 1336 224 -> 1345 210
630: 1345 210 -> 1341 196
631: 1341 196 -> 1338 172
632: 1338 172 -> 1340 153
633: 1340 153 -> 1349 132
634: 1349 132 -> 1345 109
635: 1345 109 -> 1347  80
636: 1347  80 -> 1356  59
637: 1356  59 -> 1359  42
638: 1359  42 -> 1356  34
639: 1356  34 -> 1341  29
640: 1341  29 -> 1316  29
641: 1316  29 -> 1276  29
642: 1276  29 -> 1221  29
643: 1221  29 -> 1177  32
644: 1177  32 -> 1143  31
645: 1143  31 -> 1118  24
646: 1118  24 -> 1084  23
647: 1084  23 -> 1045  31
648: 1045  31 -> 1011  29
649: 1011  29 -> 991  26
650: 991  26 -> 972  19
651: 972  19 -> 953  22
652: 953  22 -> 934  31
653: 934  31 -> 909  31
654: 909  31 -> 872  29
655: 872  29 -> 822  29
656: 822  29 -> 775  31
657: 775  31 -> 742  32
658: 742  32 -> 713  37
659: 713  37 -> 683  36
660: 683  36 -> 650  40
661: 650  40 -> 619  35
662: 619  35 -> 577  34
663: 577  34 -> 547  32
664: 547  32 -> 505  32
665: 505  32 -> 455  25
666: 455  25 -> 401  23
667: 401  23 -> 346  22
668: 346  22 -> 296  31
669: 296  31 -> 240  32
670: 240  32 -> 172  31
671: 172  31 ->  91  31
672:  91  31 ->  14  25

user16991

Posted 2014-12-06T07:27:26.907

Reputation:

A second answer seems to be the only solution. Your path checks out ok, with an average acceleration of 12.6. – Logic Knight – 2015-01-21T05:34:53.867

1

Sunday Driver, Python 2, 3242

Minified code = 2382 bytes

Performance: city=86 obstacles=46 racetrack=188 gauntlet=1092

Here is my proof-of-concept program that was to demonstrate that a solution was possible. It needs some optimisation and better golfing.

Operation

  • Create a data structure of distance rings out from the destination (simple A* derivative, like flood fill)

  • Find the shorted series of straight lines to destination that do not cross non-track pixels.

  • For each straight line, accelerate and brake to minimise the turns taken.

Golfed (minified) Code

import pygame as P,sys,random
Z=255
I=int
R=range

X=sys.argv[1]
pic=P.image.load(X)
show=P.display.flip
W,H=pic.get_size()
M=P.display.set_mode((W,H))
M.blit(pic,(0,0))
show()
U=complex
ORTH=[U(-1,0),U(1,0),U(0,-1),U(0,1)]
def draw(line,O):
 for p in line:
  M.set_at((I(p.real),I(p.imag)),O)
def plot(p,O):
 M.set_at((I(p.real),I(p.imag)),O)
def J(p):
 return abs(I(p.real))+abs(I(p.imag))
locs=[(x,y)for x in R(W)for y in R(H)]
n={}
for p in locs:
 O=tuple(M.get_at(p))[:3]
 if O not in n:
  n[O]=set()
 n[O].add(U(p[0],p[1]))
z=set()
for c in n:
 if c[0]==c[1]==c[2]and 30<=c[0]<=220 or c==(0,0,0)or c==(Z,Z,0):
  z|=n[c]
first=next(iter(n[(0,0,0)]))
ring=set([first])
s={0:ring}
g={first:0}
T=set()
G=0
done=0
while not done:
 G+=1
 T|=ring
 D=set()
 for dot in ring:
  for K in[dot+diff for diff in ORTH]:
   if K in n[(Z,Z,0)]:
    V=K;done=1
   if K in z and K not in T:
    D.add(K);g[K]=G
 ring=D
 s[G]=ring
def A(p1,p2):
 x1,y1=I(p1.real),I(p1.imag)
 x2,y2=I(p2.real),I(p2.imag)
 dx=x2-x1
 dy=y2-y1
 line=[]
 if abs(dx)>abs(dy):
  m=1.0*dy/dx
  line=[U(x,I(m*(x-x1)+y1+.5))for x in R(x1,x2,cmp(dx,0))]
 else:
  m=1.0*dx/dy
  line=[U(I(m*(y-y1)+x1+.5),y)for y in R(y1,y2,cmp(dy,0))]
 return line+[U(x2,y2)]
def f(p1,p2):
 return all(p in z for p in A(p1,p2))
def a(j,G):
 l=list(s[G])
 for F in R(150):
  w=random.choice(l)
  if f(j,w):
   return w
 return None
def d(j):
 u=g[j]
 E=k=0
 r=j
 while 1:
  w=a(j,k)
  if w:
   u=k;r=w
  else:
   E=k
  k=(u+E)/2
  if k==u or k==E:
   break
 return r
def h(p1,p2):
 if abs(p2-p1)<9:
  return p2
 line=A(p1,p2)
 tries=min(20,len(line)/2)
 test=[line[-i]for i in R(1,tries)]
 q=[(p,d(p))for p in test]
 rank=[(abs(p3-p)+abs(p-p1),p)for p,p3 in q]
 return max(rank)[1]
o=V
path=[V]
while g[o]>0:
 o=d(o)
 if o not in n[(0,0,0)]:
  o=h(path[-1],o)
 path.append(o)
 if o in n[(0,0,0)]:
  break
def t(line,N):
 v=[]
 S=len(line)/2+2
 base=i=0
 b=0
 while i<len(line):
  C=(i<S)
  Q=line[i]-line[base]
  accel=Q-N
  L=(J(accel)<=15)
  if L:
   b=1
  if C:
   if b and not L:
    i-=1;v.append(i);N=Q;base=i;b=0
  else:
   if b and J(Q)>13:
    v.append(i);N=Q;base=i;b=0
  i+=1
 v.append(i-1)
 return v,Q
turns=0
vel=U(0,0)
for V,stop in zip(path,path[1:]):
 line=A(V,stop)
 Y,vel=t(line,vel)
 turns+=len(Y)
 draw(line,(Z,Z,Z))
 plot(line[0],(0,0,Z))
 for m in Y:
  plot(line[m],(0,0,Z))
B=X.replace('.','%u.'%turns)
P.image.save(M,B)

Examples

city

racetrack

obstacles

gauntlet

Logic Knight

Posted 2014-12-06T07:27:26.907

Reputation: 6 622

At last something that does not involve brute force pathing. I'm pretty sure you could gain both in aesthetics and efficiency by smoothing the corners with a simple post-optimizer. – None – 2015-01-30T10:51:42.920

Tempting, but it wouldn't look good to do too well on my own competition. I just thought I would post my proof of concept code as an alternative approach. – Logic Knight – 2015-01-30T11:09:34.553

Come on, don't be shy :) – None – 2015-01-30T17:03:11.320

My spare time is going into a new challenge. I think you might like it. Should be posted in the next few days. – Logic Knight – 2015-01-30T17:25:57.060

I should get started on the C++ racer in the mean time, then – None – 2015-01-30T18:10:27.023