//big thanks to the coding train for teaching me how to make Neural Networks: https://www.youtube.com/playlist?list=PLRqwX-V7Uu6aCibgK1PTWWu9by6XFdCfh var popu; var unit = 0; var off = new p5.Vector (0,0); var speed = 5; var frame = 0; var drawWait = 0; var drawMax = 1; var doDraw = true; var sLen = 1*(1/3); var spdElt, mutElt, lenElt; var generation = 0; var bestTime = Infinity; var bestBrain = null; var bestFit = 0; var prevRes = 0; var obstacles2 = [new Obstacle(0, 0.175, 0.75, 0.05), new Obstacle(0.9, 0, 0.05, 0.05), new Obstacle(0.95, 0.05, 0.05, 0.05), new Obstacle(0.95, 0.3, 0.05, 0.05), new Obstacle(0.9, 0.35, 0.05, 0.05), new Obstacle(0.25, 0.4, 0.65, 0.05), new Obstacle(0.05, 0.225, 0.05, 0.05), new Obstacle(0.0, 0.275, 0.05, 0.05), new Obstacle(0.0, 0.525, 0.05, 0.05), new Obstacle(0.05, 0.575, 0.05, 0.05), new Obstacle(0.1, 0.625, 0.65, 0.05), new Obstacle(0.75, 0.675, 0.05, 0.325), new Obstacle(0.9, 0.45, 0.05, 0.05), new Obstacle(0.95, 0.5, 0.05, 0.05)]; var obstacles = [new Wall(0, 0.2, 0.75, 0), new Wall(0.875, 0, 0.125, 0.125)/*, new Wall(0.95, 0.05, 0.05, 0.05)*/, new Wall(1, 0.3, -0.125, 0.125)/*, new Wall(0.9, 0.35, 0.05, 0.05)*/, new Wall(0.25, 0.425, 0.625, 0), new Wall(0.125, 0.2, -0.125, 0.125)/*, new Obstacle(0.0, 0.275, 0.05, 0.05)*/, new Wall(0.0, 0.525, 0.125, 0.125)/*, new Obstacle(0.05, 0.575, 0.05, 0.05)*/, new Wall(0.125, 0.65, 0.65, 0.0), new Wall(0.775, 0.65, 0, 0.35), new Wall(0.875, 0.425, 0.125, 0.125)/*, new Obstacle(0.95, 0.5, 0.05, 0.05)*/]; var checkpoints = [ {x: 0, y: 0}, {x: 0.9, y: 0.2}, {x: 0.1, y: 0.425}, {x: 0.9, y: 0.9} ]; function setup () { createCanvas (windowWidth, windowHeight); unit = min (width, height); if (width > height) { this.off = new p5.Vector ((width-height)/2, 0); } else { this.off = new p5.Vector (0, (height-width)/2); } popu = new Population (10); spdElt = document.getElementById("spd"); mutElt = document.getElementById("mut"); lenElt = document.getElementById("len"); randomSeed (); } function draw () { drawMax = spdElt.value; document.getElementById ("spdL").innerText = spdElt.value; popu.mutRate = mutElt.value; document.getElementById ("mutL").innerText = mutElt.value; sLen = lenElt.value*(1/3); document.getElementById ("lenL").innerText = lenElt.value; /*drawWait++; doDraw=false; if(drawWait >= drawMax) { drawWait = 0; doDraw = true; }*/ doDraw=false; if(drawMax == 0) popu.update(); for(var k = 0; k height) { this.off = new p5.Vector ((width-height)/2, 0); } else { this.off = new p5.Vector (0, (height-width)/2); } } function NeuralNetwork (numI, numH, numO) { this.input_layer = []; this.hidden_layer = []; this.output_layer = []; for (var i=0; i < numI; i++) { this.input_layer[i] = 0; } for (var i=0; i < numH ; i++) { this.hidden_layer [i] = new Perceptron(numI); } for (var i=0; i < numO ; i++) { this.output_layer [i] = new Perceptron(numH); } } function Perceptron(n) { this.weights = []; this.lr = 0.1; for (var i=0; i < n ; i++) { this.weights [i] = random (-1, 1); } this.guess = function (inputs) { var sum = 0; for (var i=0; i < this.weights.length; i++) { sum += inputs [i] * this.weights [i]; } if(Math.tanh)return Math.tanh(sum); return sum; } } function Population (size) { this.size = size; this.cars = []; this.index = 0; this.mutRate = 0.05; for (var i=0; i < this.size; i++) { this.cars [i] = new Car (new NeuralNetwork (7, 10, 2)); } this.update = function () { var outputs, results; var dead = 0; this.index=0; for (var j=0; j < this.size; j++) { this.index = j; if(!this.cars[this.index].disabled) { try { brain = this.cars [this.index].brain; //console.log (this.cars [this.index].angle); var ray1 = new RayCast (this.cars [this.index].pos.x, this.cars [this.index].pos.y, this.cars [this.index].angle-45, sLen*unit); var ray2 = new RayCast (this.cars [this.index].pos.x, this.cars [this.index].pos.y, this.cars [this.index].angle, sLen*unit); var ray3 = new RayCast (this.cars [this.index].pos.x, this.cars [this.index].pos.y, this.cars [this.index].angle+45, sLen*unit) var ray4 = new RayCast (this.cars [this.index].pos.x, this.cars [this.index].pos.y, this.cars [this.index].angle+22.5, sLen*unit) var ray5 = new RayCast (this.cars [this.index].pos.x, this.cars [this.index].pos.y, this.cars [this.index].angle-22.5, sLen*unit) brain.input_layer = [ray1.dist/(sLen*unit/2)-1, ray2.dist/(sLen/2*unit)-1, ray3.dist/(sLen/2*unit)-1, ray4.dist/(sLen/2*unit)-1, ray5.dist/(sLen/2*unit)-1, prevRes, this.cars[this.index].vel.mag()]; outputs = []; for (var i=0; i < brain.hidden_layer.length; i++) { outputs [i] = brain.hidden_layer[i].guess (brain.input_layer); } results = []; for (var i=0; i < brain.output_layer.length; i++) { results [i] = brain.output_layer[i].guess (outputs); } prevRes = results[0]; //if(ray1.dist < 0.01 || ray2.dist < 0.01 || ray3.dist < 0.01) this.cars[this.index].disabled=true; this.cars[this.index].angle += results[1]*15/speed; if (this.cars[this.index].update (p5.Vector.fromAngle(radians (this.cars [this.index].angle)).mult (results [0]))) { //this.index++ this.cars[this.index].disabled = true; //dead++; ; } } catch { } } if (this.index != size && doDraw) { this.cars[this.index].show(); } /*else { this.index = 0; this.reproduce (); generation++; }*/ if(!this.cars[this.index].disabled) this.cars [this.index].brain.input_layer = brain.input_layer; else dead++; //this.index += 1; } if(dead==this.size) { this.reproduce (); generation++; frame=0; } return [outputs, results]; } this.reproduce = function () { var fitnesses = []; var maxFit = 0; for (var i=0; i < this.cars.length; i++) { var curPass = this.cars[i].checkPassed; var maxDist = distVec(checkpoints[curPass-1].x, checkpoints[curPass-1].y, checkpoints[curPass].x, checkpoints[curPass].y); var curDist = 1-distVec(this.cars [i].pos.x, this.cars [i].pos.y, checkpoints[curPass].x, checkpoints[curPass].y)/maxDist; fitnesses[i] = Math.pow((Math.pow(curDist*100, 2)/10000+this.cars[i].checkPassed-1)*10, 3)/10; if (this.cars[i].passed) { fitnesses[i] += Math.pow(80000/this.cars[i].time, 3); } if (fitnesses[i] > maxFit) maxFit = fitnesses[i]; if(fitnesses[i] > bestFit) { bestFit = fitnesses[i]; bestBrain = this.cars[i].brain; } } //regulate fitness between 1 and 0 for (var i=0; i < this.cars.length; i++) { fitnesses[i] /= maxFit; } var pool = []; for (var i=0; i < this.cars.length; i++) { for (var j=0; j600*this.checkPassed) this.disabled = true; var prevPos = this.pos.copy (); this.vel.add (acc.mult (0.001/speed)); this.pos.add (this.vel); this.vel.mult (0.95); if (this.vel.mag > this.maxSpeed) this.vel.setMag (this.maxSpeed) var collide = false; for (var i=0; i obs.x && this.pos.y-this.r < obs.y+obs.h && this.pos.y+this.r > obs.y) { collide = true; break; }*/ } if(checkpoints[this.checkPassed].y < this.pos.y) { if(this.checkPassed < checkpoints.length-1) this.checkPassed++; else { this.passed = true; if (this.time < bestTime) { bestTime = this.time; bestBrain = JSON.stringify(this.brain); } return true; } } return (this.borders() || (this.pos.x > 0.95 && this.pos.y > 0.95) || prevPos.dist (this.pos) < 0.3/speed/unit || collide); } this.show = function () { fill (0,0,200); stroke (0,0,255); rect (off.x+(this.pos.x-this.r)*unit, off.y+(this.pos.y-this.r)*unit, this.r*2*unit, this.r*2*unit) } this.borders = function () { return (this.pos.x < 0 || this.pos.y < 0 || this.pos.x > 1 || this.pos.y > 1); //console.log (popu.index) } } function RayCast (x, y, dir, length) { //this.pos = new p5.Vector(x*unit, y*unit); this.vel = new p5.Vector.fromAngle (radians(dir)); this.pos = new p5.Vector(x, y); this.pos2 = this.vel.setMag(length); /*for(var i=0; i obs.x && (this.pos.y-1)/unit < obs.y+obs.h && (this.pos.y+1)/unit > obs.y) { collide = true; break; } } if (this.pos.x < 0 || this.pos.y < 0 || this.pos.x > unit || this.pos.y > unit || collide) { this.dist = i; break; } }*/ var best = null; this.dist = length; obstaclesmax = obstacles.concat([new Wall(0,0,1,0), new Wall(1,0,0,1), new Wall(0,1,1,0), new Wall(0,0,0,1)]); for (var j=0; j0 && t<1 && u>0 && u<1) { var pd = createVector(); pd.x = x1+t*(x2-x1); pd.y = y1+t*(y2-y1); var d = distVec(this.pos.x, this.pos.y, pd.x, pd.y); if(d*unit < this.dist) { this.dist = d*unit; best = pd; } }*/ /*if(t>0 && t<1 && u>0 && u<1) var tmsg = t;//JSON.stringify(obs[obstacles.length-1]); if(j==0) text ("object: " + tmsg, 0, 48);*/ } if (this.dist == length) { this.dist = length; stroke (0, 255, 0) if(doDraw) line (off.x+x*unit, off.y+y*unit, off.x+x*unit+this.vel.setMag(this.dist).x, off.y+y*unit+this.vel.setMag(this.dist).y); } else { stroke (255, 0, 0) if(doDraw) line (off.x+x*unit, off.y+y*unit, off.x+x*unit+this.vel.setMag(this.dist).x, off.y+y*unit+this.vel.setMag(this.dist).y); } } function lineCollide(l1, l2) { const x1 = l1.x; const y1 = l1.y; const x2 = l1.x+l1.w; const y2 = l1.y+l1.h; const x3 = l2.x; const y3 = l2.y; const x4 = l2.x+l2.w; const y4 = l2.y+l2.h; den = (x1-x2)*(y3-y4)-(y1 - y2)*(x3-x4); if(den == 0) return; const t = ((x1-x3)*(y3-y4)-(y1 - y3)*(x3-x4))/den; const u = -((x1-x2)*(y1-y3)-(y1 - y2)*(x1-x3))/den; if(t>0 && t<1 && u>0 && u<1) { var pd = createVector(); pd.x = x1+t*(x2-x1); pd.y = y1+t*(y2-y1); return pd; } } function Obstacle (x, y, w, h) { this.x = x; this.y = y; this.w = w; this.h = h; this.show = function () { stroke (255); fill (127); rect (off.x+this.x*unit, off.y+this.y*unit, this.w*unit, this.h*unit); } } function Wall (x, y, w, h) { this.x = x; this.y = y; this.w = w; this.h = h; this.show = function () { stroke (255); fill (127); line (off.x+this.x*unit, off.y+this.y*unit, off.x+(this.x+this.w)*unit, off.y+(this.y+this.h)*unit); } } function Line (x, y, x2, y2) { this.x = x; this.y = y; this.x2 = x2; this.y2 = y2; this.show = function () { stroke (255); fill (127); line (off.x+this.x*unit, off.y+this.y*unit, off.x+this.x2*unit, off.y+this.y2*unit); } } function distVec(x1, y1, x2, y2) { return Math.sqrt(Math.pow(x1-x2, 2) + Math.pow(y1-y2, 2)) } function keyPressed() { if(keyCode == ENTER) { popu.reproduce(); generation++; } if(key == "R" && bestBrain) { for(var i=0; i

Self Driving Cars