LiorB-D · July 15, 2022 11:38
diff --git a/index.html b/index.html
 <head>
  <!-- Load in Tensorflow, P5, and our Car Class-->
  <script src="https://cdn.jsdelivr.net/npm/@tensorflow/[email protected]/dist/tf.min.js"></script>
  <script
    src="https://cdnjs.cloudflare.com/ajax/libs/p5.js/1.1.9/p5.min.js"
    integrity="sha512-WIklPM6qPCIp6d3fSSr90j+1unQHUOoWDS4sdTiR8gxUTnyZ8S2Mr8e10sKKJ/bhJgpAa/qG068RDkg6fIlNFA=="
    crossorigin="anonymous"
  ></script>
  <script src="car.js"></script>

  <script>
    // Screen dimensions
    const screenHt = 650;
    const screenWd = 800;
    const innerDi = 300; // Diameter of Inner Circle
    const outerDi = 750; // Diameter of Outer Circle
    const carDi = 20; // Diameter of each car

    let frameCount = 0;

    let currGen = 1;
    let topFitnessScore = 0; // Highest fitness score of any generation

    //Obstacles can be added/removed without any additional implementation
    let obsts = [
      { x: screenWd / 2, y: screenHt / 4, di: 40 },
      { x: screenWd / 2, y: (3 * screenHt) / 4, di: 40 },
    ];

    // UI elements
    let nextGenBtn, mrateInput, mrateLbl;
    let infoLbl;

    let cars = [];

    function setup() {
      createCanvas(screenWd + 20 + 200, screenHt + 20);
      frameRate(60); // Edit Frame Rate to change speed of simulation

      // Setup UI elements with P5
      nextGenBtn = createButton("Next Generation");
      nextGenBtn.position(screenWd + 30, 110);

      mrateInput = createInput("0.15");
      mrateInput.position(screenWd + 30, 80);

      mrateLbl = createElement("p", "Mutation Rate:");
      mrateLbl.position(screenWd + 30, 45);

      infoLbl = createElement(
        "p",
        "Gen: " + currGen + ", Top Score: " + topFitnessScore
      );
      infoLbl.position(screenWd + 30, 150);
      // Add 20 new cars
      for (let i = 0; i < 20; i++) {
        cars.push(new Car(outerDi, innerDi, screenHt, screenWd, carDi, obsts));
      }
    }

    function draw() {
      // Looped based on framerate

      //Draw Driving Map
      strokeWeight(1);
      fill(50);
      stroke("black");
      rect(0, 0, screenWd, screenHt);

      //Outer Circle
      fill(210);
      ellipse(screenWd / 2, screenHt / 2, outerDi, outerDi / 1.5);

      fill("green");
      strokeWeight(3);
      //Obstacles
      obsts.forEach((obs) => {
        ellipse(obs.x, obs.y, obs.di);
      });
      //Inner circle
      ellipse(screenWd / 2, screenHt / 2, innerDi, innerDi / 2);

      //Draw Finish Line
      stroke("red");
      strokeWeight(4);
      line(
        screenWd / 2 - outerDi / 2,
        screenHt / 2,
        screenWd / 2 - innerDi / 2,
        screenHt / 2
      );

      cars.forEach((car) => {
        //Draw car

        //Draw Wheels
        fill("black");
        strokeWeight(0);
        ellipse(
          car.x + (carDi / 2) * cos(car.angle + PI / 4),
          car.y + (carDi / 2) * sin(car.angle + PI / 4),
          carDi / 2.5
        );
        ellipse(
          car.x + (carDi / 2) * cos(car.angle - PI / 4),
          car.y + (carDi / 2) * sin(car.angle - PI / 4),
          carDi / 2.5
        );
        ellipse(
          car.x + (carDi / 2) * cos(car.angle + 0.75 * PI),
          car.y + (carDi / 2) * sin(car.angle + 0.75 * PI),
          carDi / 2.5
        );
        ellipse(
          car.x + (carDi / 2) * cos(car.angle - 0.75 * PI),
          car.y + (carDi / 2) * sin(car.angle - 0.75 * PI),
          carDi / 2.5
        );

        //Draw Car Base
        fill("blue");
        stroke("black");
        strokeWeight(1);
        ellipse(car.x, car.y, carDi);

        //Draw Eye
        fill("yellow");
        ellipse(car.eyeX, car.eyeY, carDi / 3);

        // Car Behavior

        if (car.dead == false && car.ptInObst(car.x, car.y) == false) {
          // Make sure car is alive
          car.updatePos();

          //Make Decision

          //Get obstacle detection distances
          let inputs = tf.tensor2d([car.getCollisDist()]);

          // Predict the value
          let decArr = car.model.predict(inputs).dataSync();

          //Adjust angle and velocity based on outputs
          car.v += 0.1 * decArr[0]; // Accelerate
          car.v -= 0.2 * decArr[1]; // Brake
          car.angle -= 0.1 * decArr[2]; // Turn Left
          car.angle += 0.1 * decArr[3]; // Turn Right
        } else {
          car.dead = true; // Kill the car if it is in obstacle
        }
      });

      frameCount += 1;
      // If 200 frames has passed plus a scaled of the highestScore, then start next generation
      if (frameCount > 200 + topFitnessScore * 3) {
        startNewGen();
        frameCount = 0;
      } else if (frameCount > 80) { // Check for inactivity after 80 frames
        let stillActive = false;
        // Make sure atleast one car has made progress without dying
        cars.forEach((car) => {
          if (car.fitness > 3 && car.dead == false) {
            stillActive = true;
          }
        });
        if (stillActive == false) {
          startNewGen();
          frameCount = 0;
        }
      }
    }

    function endGen() {
      let totalFitness = 0;
      bestOfGen = 0;

      // Calculate total fitness and find highest performing car
      cars.forEach((car, ind) => {
        car.dead = true;
        totalFitness += car.fitness;
        if (car.fitness > bestOfGen) {
          bestOfGen = car.fitness;
        }
      });

      //Update highest score if applicable
      if (bestOfGen > topFitnessScore) {
        topFitnessScore = bestOfGen;
      }

      return totalFitness;
    }

    function selectRandParent(tFit) {
      let randVal = tFit * random(); // Random value between (0, tFit)
      let count = 0; // Cumulitive Fitness

      for (let i = 0; i < cars.length; i++) {
        count += cars[i].fitness;
        if (randVal <= count) {
          return i;
        }
      }
    }

    function startNewGen() {
      totalFitness = endGen();
      let newCarArr = []; // The next generation

      mutationRate = mrateInput.value() // Pull mutation rate

      //Get parent weights
      let parentWeights = [];
      cars.forEach((c) => {
        parentWeights.push(c.copyWeights());
      });

      //Create next generation
      for (let i = 0; i < 10; i++) {
        // Initialize Child
        const child = new Car(
          outerDi,
          innerDi,
          screenHt,
          screenWd,
          carDi,
          obsts
        );

        let currWeights = child.model.getWeights();
        let newWeights = [];

        for (let i = 0; i < currWeights.length; i++) {
          // Iterate through each layer
          //Get Current Weight information
          let currVals = currWeights[i].dataSync(); // Array format
          let shape = currWeights[i].shape;

          for (let j = 0; j < currVals.length; j++) {
            // Iterate through each weight
            // Select trait from random parent
            currVals[j] = parentWeights[selectRandParent(totalFitness)][i][j];
            // Assign random value to weight from Gaussian Distribution(Function from P5)
            if(random() < mutationRate) {
                currVals[j] += randomGaussian()
            }
          }
          let newTens = tf.tensor(currVals, shape);
          newWeights[i] = newTens;
        }

        child.model.setWeights(newWeights);

        newCarArr.push(child);
      }

      cars = newCarArr;
      // Increment Generation and Update Label
      currGen += 1;
      infoLbl.elt.innerText =
        "Gen: " + currGen + ", Top Score: " + topFitnessScore;
    }
  </script>

  <title>Learning to Drive w/ Genetic Algorithm</title>
 </head>
	<head>
	<!-- Load in Tensorflow, P5, and our Car Class-->
	<script src="https://cdn.jsdelivr.net/npm/@tensorflow/[email protected]/dist/tf.min.js"></script>
	<script
	src="https://cdnjs.cloudflare.com/ajax/libs/p5.js/1.1.9/p5.min.js"
	integrity="sha512-WIklPM6qPCIp6d3fSSr90j+1unQHUOoWDS4sdTiR8gxUTnyZ8S2Mr8e10sKKJ/bhJgpAa/qG068RDkg6fIlNFA=="
	crossorigin="anonymous"
	></script>
	<script src="car.js"></script>

	<script>
	// Screen dimensions
	const screenHt = 650;
	const screenWd = 800;
	const innerDi = 300; // Diameter of Inner Circle
	const outerDi = 750; // Diameter of Outer Circle
	const carDi = 20; // Diameter of each car

	let frameCount = 0;

	let currGen = 1;
	let topFitnessScore = 0; // Highest fitness score of any generation

	//Obstacles can be added/removed without any additional implementation
	let obsts = [
	{ x: screenWd / 2, y: screenHt / 4, di: 40 },
	{ x: screenWd / 2, y: (3 * screenHt) / 4, di: 40 },
	];

	// UI elements
	let nextGenBtn, mrateInput, mrateLbl;
	let infoLbl;

	let cars = [];

	function setup() {
	createCanvas(screenWd + 20 + 200, screenHt + 20);
	frameRate(60); // Edit Frame Rate to change speed of simulation

	// Setup UI elements with P5
	nextGenBtn = createButton("Next Generation");
	nextGenBtn.position(screenWd + 30, 110);

	mrateInput = createInput("0.15");
	mrateInput.position(screenWd + 30, 80);

	mrateLbl = createElement("p", "Mutation Rate:");
	mrateLbl.position(screenWd + 30, 45);

	infoLbl = createElement(
	"p",
	"Gen: " + currGen + ", Top Score: " + topFitnessScore
	);
	infoLbl.position(screenWd + 30, 150);
	// Add 20 new cars
	for (let i = 0; i < 20; i++) {
	cars.push(new Car(outerDi, innerDi, screenHt, screenWd, carDi, obsts));
	}
	}

	function draw() {
	// Looped based on framerate

	//Draw Driving Map
	strokeWeight(1);
	fill(50);
	stroke("black");
	rect(0, 0, screenWd, screenHt);

	//Outer Circle
	fill(210);
	ellipse(screenWd / 2, screenHt / 2, outerDi, outerDi / 1.5);

	fill("green");
	strokeWeight(3);
	//Obstacles
	obsts.forEach((obs) => {
	ellipse(obs.x, obs.y, obs.di);
	});
	//Inner circle
	ellipse(screenWd / 2, screenHt / 2, innerDi, innerDi / 2);

	//Draw Finish Line
	stroke("red");
	strokeWeight(4);
	line(
	screenWd / 2 - outerDi / 2,
	screenHt / 2,
	screenWd / 2 - innerDi / 2,
	screenHt / 2
	);

	cars.forEach((car) => {
	//Draw car

	//Draw Wheels
	fill("black");
	strokeWeight(0);
	ellipse(
	car.x + (carDi / 2) * cos(car.angle + PI / 4),
	car.y + (carDi / 2) * sin(car.angle + PI / 4),
	carDi / 2.5
	);
	ellipse(
	car.x + (carDi / 2) * cos(car.angle - PI / 4),
	car.y + (carDi / 2) * sin(car.angle - PI / 4),
	carDi / 2.5
	);
	ellipse(
	car.x + (carDi / 2) * cos(car.angle + 0.75 * PI),
	car.y + (carDi / 2) * sin(car.angle + 0.75 * PI),
	carDi / 2.5
	);
	ellipse(
	car.x + (carDi / 2) * cos(car.angle - 0.75 * PI),
	car.y + (carDi / 2) * sin(car.angle - 0.75 * PI),
	carDi / 2.5
	);

	//Draw Car Base
	fill("blue");
	stroke("black");
	strokeWeight(1);
	ellipse(car.x, car.y, carDi);

	//Draw Eye
	fill("yellow");
	ellipse(car.eyeX, car.eyeY, carDi / 3);

	// Car Behavior

	if (car.dead == false && car.ptInObst(car.x, car.y) == false) {
	// Make sure car is alive
	car.updatePos();

	//Make Decision

	//Get obstacle detection distances
	let inputs = tf.tensor2d([car.getCollisDist()]);

	// Predict the value
	let decArr = car.model.predict(inputs).dataSync();

	//Adjust angle and velocity based on outputs
	car.v += 0.1 * decArr[0]; // Accelerate
	car.v -= 0.2 * decArr[1]; // Brake
	car.angle -= 0.1 * decArr[2]; // Turn Left
	car.angle += 0.1 * decArr[3]; // Turn Right
	} else {
	car.dead = true; // Kill the car if it is in obstacle
	}
	});

	frameCount += 1;
	// If 200 frames has passed plus a scaled of the highestScore, then start next generation
	if (frameCount > 200 + topFitnessScore * 3) {
	startNewGen();
	frameCount = 0;
	} else if (frameCount > 80) { // Check for inactivity after 80 frames
	let stillActive = false;
	// Make sure atleast one car has made progress without dying
	cars.forEach((car) => {
	if (car.fitness > 3 && car.dead == false) {
	stillActive = true;
	}
	});
	if (stillActive == false) {
	startNewGen();
	frameCount = 0;
	}
	}
	}

	function endGen() {
	let totalFitness = 0;
	bestOfGen = 0;

	// Calculate total fitness and find highest performing car
	cars.forEach((car, ind) => {
	car.dead = true;
	totalFitness += car.fitness;
	if (car.fitness > bestOfGen) {
	bestOfGen = car.fitness;
	}
	});

	//Update highest score if applicable
	if (bestOfGen > topFitnessScore) {
	topFitnessScore = bestOfGen;
	}

	return totalFitness;
	}

	function selectRandParent(tFit) {
	let randVal = tFit * random(); // Random value between (0, tFit)
	let count = 0; // Cumulitive Fitness

	for (let i = 0; i < cars.length; i++) {
	count += cars[i].fitness;
	if (randVal <= count) {
	return i;
	}
	}
	}

	function startNewGen() {
	totalFitness = endGen();
	let newCarArr = []; // The next generation

	mutationRate = mrateInput.value() // Pull mutation rate

	//Get parent weights
	let parentWeights = [];
	cars.forEach((c) => {
	parentWeights.push(c.copyWeights());
	});

	//Create next generation
	for (let i = 0; i < 10; i++) {
	// Initialize Child
	const child = new Car(
	outerDi,
	innerDi,
	screenHt,
	screenWd,
	carDi,
	obsts
	);

	let currWeights = child.model.getWeights();
	let newWeights = [];

	for (let i = 0; i < currWeights.length; i++) {
	// Iterate through each layer
	//Get Current Weight information
	let currVals = currWeights[i].dataSync(); // Array format
	let shape = currWeights[i].shape;

	for (let j = 0; j < currVals.length; j++) {
	// Iterate through each weight
	// Select trait from random parent
	currVals[j] = parentWeights[selectRandParent(totalFitness)][i][j];
	// Assign random value to weight from Gaussian Distribution(Function from P5)
	if(random() < mutationRate) {
	currVals[j] += randomGaussian()
	}
	}
	let newTens = tf.tensor(currVals, shape);
	newWeights[i] = newTens;
	}

	child.model.setWeights(newWeights);

	newCarArr.push(child);
	}

	cars = newCarArr;
	// Increment Generation and Update Label
	currGen += 1;
	infoLbl.elt.innerText =
	"Gen: " + currGen + ", Top Score: " + topFitnessScore;
	}
	</script>

	<title>Learning to Drive w/ Genetic Algorithm</title>
	</head>