Lecture 7 code

7. Visualizing neural networks/car.js

@@ -0,0 +1,141 @@
+class Car{
+    constructor(x,y,width,height,controlType,maxSpeed=3){
+        this.x=x;
+        this.y=y;
+        this.width=width;
+        this.height=height;
+
+        this.speed=0;
+        this.acceleration=0.2;
+        this.maxSpeed=maxSpeed;
+        this.friction=0.05;
+        this.angle=0;
+        this.damaged=false;
+
+        this.useBrain=controlType=="AI";
+
+        if(controlType!="DUMMY"){
+            this.sensor=new Sensor(this);
+            this.brain=new NeuralNetwork(
+                [this.sensor.rayCount,6,4]
+            );
+        }
+        this.controls=new Controls(controlType);
+    }
+
+    update(roadBorders,traffic){
+        if(!this.damaged){
+            this.#move();
+            this.polygon=this.#createPolygon();
+            this.damaged=this.#assessDamage(roadBorders,traffic);
+        }
+        if(this.sensor){
+            this.sensor.update(roadBorders,traffic);
+            const offsets=this.sensor.readings.map(
+                s=>s==null?0:1-s.offset
+            );
+            const outputs=NeuralNetwork.feedForward(offsets,this.brain);
+
+            if(this.useBrain){
+                this.controls.forward=outputs[0];
+                this.controls.left=outputs[1];
+                this.controls.right=outputs[2];
+                this.controls.reverse=outputs[3];
+            }
+        }
+    }
+
+    #assessDamage(roadBorders,traffic){
+        for(let i=0;i<roadBorders.length;i++){
+            if(polysIntersect(this.polygon,roadBorders[i])){
+                return true;
+            }
+        }
+        for(let i=0;i<traffic.length;i++){
+            if(polysIntersect(this.polygon,traffic[i].polygon)){
+                return true;
+            }
+        }
+        return false;
+    }
+
+    #createPolygon(){
+        const points=[];
+        const rad=Math.hypot(this.width,this.height)/2;
+        const alpha=Math.atan2(this.width,this.height);
+        points.push({
+            x:this.x-Math.sin(this.angle-alpha)*rad,
+            y:this.y-Math.cos(this.angle-alpha)*rad
+        });
+        points.push({
+            x:this.x-Math.sin(this.angle+alpha)*rad,
+            y:this.y-Math.cos(this.angle+alpha)*rad
+        });
+        points.push({
+            x:this.x-Math.sin(Math.PI+this.angle-alpha)*rad,
+            y:this.y-Math.cos(Math.PI+this.angle-alpha)*rad
+        });
+        points.push({
+            x:this.x-Math.sin(Math.PI+this.angle+alpha)*rad,
+            y:this.y-Math.cos(Math.PI+this.angle+alpha)*rad
+        });
+        return points;
+    }
+
+    #move(){
+        if(this.controls.forward){
+            this.speed+=this.acceleration;
+        }
+        if(this.controls.reverse){
+            this.speed-=this.acceleration;
+        }
+
+        if(this.speed>this.maxSpeed){
+            this.speed=this.maxSpeed;
+        }
+        if(this.speed<-this.maxSpeed/2){
+            this.speed=-this.maxSpeed/2;
+        }
+
+        if(this.speed>0){
+            this.speed-=this.friction;
+        }
+        if(this.speed<0){
+            this.speed+=this.friction;
+        }
+        if(Math.abs(this.speed)<this.friction){
+            this.speed=0;
+        }
+
+        if(this.speed!=0){
+            const flip=this.speed>0?1:-1;
+            if(this.controls.left){
+                this.angle+=0.03*flip;
+            }
+            if(this.controls.right){
+                this.angle-=0.03*flip;
+            }
+        }
+
+        this.x-=Math.sin(this.angle)*this.speed;
+        this.y-=Math.cos(this.angle)*this.speed;
+    }
+
+    draw(ctx,color){
+        if(this.damaged){
+            ctx.fillStyle="gray";
+        }else{
+            ctx.fillStyle=color;
+        }
+        ctx.beginPath();
+        ctx.moveTo(this.polygon[0].x,this.polygon[0].y);
+        for(let i=1;i<this.polygon.length;i++){
+            ctx.lineTo(this.polygon[i].x,this.polygon[i].y);
+        }
+        ctx.fill();
+
+        if(this.sensor){
+            this.sensor.draw(ctx);
+        }
+    }
+}

7. Visualizing neural networks/controls.js

@@ -0,0 +1,52 @@
+class Controls{
+    constructor(type){
+        this.forward=false;
+        this.left=false;
+        this.right=false;
+        this.reverse=false;
+
+        switch(type){
+            case "KEYS":
+                this.#addKeyboardListeners();
+                break;
+            case "DUMMY":
+                this.forward=true;
+                break;
+        }
+    }
+
+    #addKeyboardListeners(){
+        document.onkeydown=(event)=>{
+            switch(event.key){
+                case "ArrowLeft":
+                    this.left=true;
+                    break;
+                case "ArrowRight":
+                    this.right=true;
+                    break;
+                case "ArrowUp":
+                    this.forward=true;
+                    break;
+                case "ArrowDown":
+                    this.reverse=true;
+                    break;
+            }
+        }
+        document.onkeyup=(event)=>{
+            switch(event.key){
+                case "ArrowLeft":
+                    this.left=false;
+                    break;
+                case "ArrowRight":
+                    this.right=false;
+                    break;
+                case "ArrowUp":
+                    this.forward=false;
+                    break;
+                case "ArrowDown":
+                    this.reverse=false;
+                    break;
+            }
+        }
+    }
+}

7. Visualizing neural networks/index.html

@@ -0,0 +1,18 @@
+<!DOCTYPE html>
+<head>
+    <title>Self-driving car - No libraries</title>
+    <link rel="stylesheet" href="style.css">
+</head>
+<body>
+    <canvas id="carCanvas"></canvas>
+    <canvas id="networkCanvas"></canvas>
+    <script src="visualizer.js"></script>
+    <script src="network.js"></script>
+    <script src="sensor.js"></script>
+    <script src="utils.js"></script>
+    <script src="road.js"></script>
+    <script src="controls.js"></script>
+    <script src="car.js"></script>
+    <script src="main.js"></script>
+</body>
+</html>

7. Visualizing neural networks/main.js

@@ -0,0 +1,40 @@
+const carCanvas=document.getElementById("carCanvas");
+carCanvas.width=200;
+const networkCanvas=document.getElementById("networkCanvas");
+networkCanvas.width=300;
+
+const carCtx = carCanvas.getContext("2d");
+const networkCtx = networkCanvas.getContext("2d");
+
+const road=new Road(carCanvas.width/2,carCanvas.width*0.9);
+const car=new Car(road.getLaneCenter(1),100,30,50,"AI");
+const traffic=[
+    new Car(road.getLaneCenter(1),-100,30,50,"DUMMY",2)
+];
+
+animate();
+
+function animate(time){
+    for(let i=0;i<traffic.length;i++){
+        traffic[i].update(road.borders,[]);
+    }
+    car.update(road.borders,traffic);
+
+    carCanvas.height=window.innerHeight;
+    networkCanvas.height=window.innerHeight;
+
+    carCtx.save();
+    carCtx.translate(0,-car.y+carCanvas.height*0.7);
+
+    road.draw(carCtx);
+    for(let i=0;i<traffic.length;i++){
+        traffic[i].draw(carCtx,"red");
+    }
+    car.draw(carCtx,"blue");
+
+    carCtx.restore();
+
+    networkCtx.lineDashOffset=-time/50;
+    Visualizer.drawNetwork(networkCtx,car.brain);
+    requestAnimationFrame(animate);
+}

7. Visualizing neural networks/network.js

@@ -0,0 +1,68 @@
+class NeuralNetwork{
+    constructor(neuronCounts){
+        this.levels=[];
+        for(let i=0;i<neuronCounts.length-1;i++){
+            this.levels.push(new Level(
+                neuronCounts[i],neuronCounts[i+1]
+            ));
+        }
+    }
+
+    static feedForward(givenInputs,network){
+        let outputs=Level.feedForward(
+            givenInputs,network.levels[0]);
+        for(let i=1;i<network.levels.length;i++){
+            outputs=Level.feedForward(
+                outputs,network.levels[i]);
+        }
+        return outputs;
+    }
+}
+
+class Level{
+    constructor(inputCount,outputCount){
+        this.inputs=new Array(inputCount);
+        this.outputs=new Array(outputCount);
+        this.biases=new Array(outputCount);
+
+        this.weights=[];
+        for(let i=0;i<inputCount;i++){
+            this.weights[i]=new Array(outputCount);
+        }
+
+        Level.#randomize(this);
+    }
+
+    static #randomize(level){
+        for(let i=0;i<level.inputs.length;i++){
+            for(let j=0;j<level.outputs.length;j++){
+                level.weights[i][j]=Math.random()*2-1;
+            }
+        }
+
+        for(let i=0;i<level.biases.length;i++){
+            level.biases[i]=Math.random()*2-1;
+        }
+    }
+
+    static feedForward(givenInputs,level){
+        for(let i=0;i<level.inputs.length;i++){
+            level.inputs[i]=givenInputs[i];
+        }
+
+        for(let i=0;i<level.outputs.length;i++){
+            let sum=0
+            for(let j=0;j<level.inputs.length;j++){
+                sum+=level.inputs[j]*level.weights[j][i];
+            }
+
+            if(sum>level.biases[i]){
+                level.outputs[i]=1;
+            }else{
+                level.outputs[i]=0;
+            } 
+        }
+
+        return level.outputs;
+    }
+}

7. Visualizing neural networks/road.js

@@ -0,0 +1,56 @@
+class Road{
+    constructor(x,width,laneCount=3){
+        this.x=x;
+        this.width=width;
+        this.laneCount=laneCount;
+
+        this.left=x-width/2;
+        this.right=x+width/2;
+
+        const infinity=1000000;
+        this.top=-infinity;
+        this.bottom=infinity;
+
+        const topLeft={x:this.left,y:this.top};
+        const topRight={x:this.right,y:this.top};
+        const bottomLeft={x:this.left,y:this.bottom};
+        const bottomRight={x:this.right,y:this.bottom};
+        this.borders=[
+            [topLeft,bottomLeft],
+            [topRight,bottomRight]
+        ];
+    }
+
+    getLaneCenter(laneIndex){
+        const laneWidth=this.width/this.laneCount;
+        return this.left+laneWidth/2+
+            Math.min(laneIndex,this.laneCount-1)*laneWidth;
+    }
+
+    draw(ctx){
+        ctx.lineWidth=5;
+        ctx.strokeStyle="white";
+
+        for(let i=1;i<=this.laneCount-1;i++){
+            const x=lerp(
+                this.left,
+                this.right,
+                i/this.laneCount
+            );
+
+            ctx.setLineDash([20,20]);
+            ctx.beginPath();
+            ctx.moveTo(x,this.top);
+            ctx.lineTo(x,this.bottom);
+            ctx.stroke();
+        }
+
+        ctx.setLineDash([]);
+        this.borders.forEach(border=>{
+            ctx.beginPath();
+            ctx.moveTo(border[0].x,border[0].y);
+            ctx.lineTo(border[1].x,border[1].y);
+            ctx.stroke();
+        });
+    }
+}