Navigate the maze

In this tutorial you build a complete program, step by step. The robot drives around the maze avoiding walls and holes, and reports anything that looks like a wall token. It is the program you ran in Your first run.

It uses the starter robot, and what you learned in Movement and Sensors.

The complete program: maze.py · maze.cpp

The field

Run it on world1.wbt, which looks like this:

Step by step

1. Setup

First, the program creates the robot, gets the wheel motors and lets them turn without limit. speeds holds the speed of each wheel, which the rest of the program changes. reported remembers where the robot has already reported a token.

Python

from controller import Robot
import struct
import numpy as np
import cv2 as cv

robot = Robot()
timeStep = 32         # Milliseconds between steps
maxVelocity = 6.28    # The maximum speed of the wheels, in radians per second

leftWheel = robot.getDevice("left wheel motor")
rightWheel = robot.getDevice("right wheel motor")
leftWheel.setPosition(float("inf"))   # Let the wheels turn without limit
rightWheel.setPosition(float("inf"))

speeds = [maxVelocity, maxVelocity]   # [left wheel speed, right wheel speed]
reported = []                         # Positions where a token has already been reported

C/C++

#include <array>
#include <cmath>
#include <cstring>
#include <vector>
#include <opencv2/opencv.hpp>
#include <webots/Robot.hpp>
#include <webots/Motor.hpp>
#include <webots/DistanceSensor.hpp>
#include <webots/Camera.hpp>
#include <webots/Emitter.hpp>
#include <webots/GPS.hpp>

using namespace webots;
using namespace std;

Robot *robot = new Robot();
const int timeStep = 32;          // Milliseconds between steps
const double maxVelocity = 6.28;  // The maximum speed of the wheels, in radians per second

Motor *leftWheel = robot->getMotor("left wheel motor");
Motor *rightWheel = robot->getMotor("right wheel motor");

double speeds[2] = {maxVelocity, maxVelocity}; // {left wheel speed, right wheel speed}
vector<array<int, 2>> reported;                // Positions where a token has already been reported

2. Sensors

Next, it gets and enables the sensors of the starter robot:

Python

leftDist = robot.getDevice("leftDist")
frontDist = robot.getDevice("frontDist")
rightDist = robot.getDevice("rightDist")
camera = robot.getDevice("camera")
colorSensor = robot.getDevice("color")
gps = robot.getDevice("gps")
emitter = robot.getDevice("emitter")  # The emitter does not need to be enabled

for sensor in [leftDist, frontDist, rightDist, camera, colorSensor, gps]:
    sensor.enable(timeStep)

C/C++

DistanceSensor *leftDist = robot->getDistanceSensor("leftDist");
DistanceSensor *frontDist = robot->getDistanceSensor("frontDist");
DistanceSensor *rightDist = robot->getDistanceSensor("rightDist");
Camera *camera = robot->getCamera("camera");
Camera *colorSensor = robot->getCamera("color");
GPS *gps = robot->getGPS("gps");
Emitter *emitter = robot->getEmitter("emitter"); // The emitter does not need to be enabled

void enableSensors() {
  leftDist->enable(timeStep);
  frontDist->enable(timeStep);
  rightDist->enable(timeStep);
  camera->enable(timeStep);
  colorSensor->enable(timeStep);
  gps->enable(timeStep);
}

3. Moves

These functions change the speeds to turn and spin. delay keeps the robot doing the same thing for a given number of milliseconds, and getBrightness reads the colour sensor as a single value from 0 (black) to 255 (white).

Python

def turnRight():
    speeds[0] = 0.6 * maxVelocity
    speeds[1] = -0.2 * maxVelocity

def turnLeft():
    speeds[0] = -0.2 * maxVelocity
    speeds[1] = 0.6 * maxVelocity

def spin():
    speeds[0] = 0.6 * maxVelocity
    speeds[1] = -0.6 * maxVelocity

def delay(ms):
    start = robot.getTime()
    while robot.step(timeStep) != -1:
        if (robot.getTime() - start) * 1000 > ms:
            break

def getBrightness():
    image = colorSensor.getImage()
    return colorSensor.imageGetGray(image, colorSensor.getWidth(), 0, 0)  # 0 (black) to 255 (white)

C/C++

void turnRight() {
  speeds[0] = 0.6 * maxVelocity;
  speeds[1] = -0.2 * maxVelocity;
}

void turnLeft() {
  speeds[0] = -0.2 * maxVelocity;
  speeds[1] = 0.6 * maxVelocity;
}

void spin() {
  speeds[0] = 0.6 * maxVelocity;
  speeds[1] = -0.6 * maxVelocity;
}

void delay(int ms) {
  double start = robot->getTime();
  while (robot->step(timeStep) != -1) {
    if ((robot->getTime() - start) * 1000 > ms)
      break;
  }
}

int getBrightness() {
  const unsigned char *image = colorSensor->getImage();
  return Camera::imageGetGray(image, colorSensor->getWidth(), 0, 0); // 0 (black) to 255 (white)
}

4. Seeing a token

This function looks for wall tokens in a camera image. Tokens are bright squares on dark walls, so it looks for bright shapes that are roughly square and do not touch the edge of the image, where the floor usually is.

Python

def seesToken(image):
    height, width = camera.getHeight(), camera.getWidth()
    image = np.frombuffer(image, np.uint8).reshape((height, width, 4))  # BGRA image for OpenCV
    gray = cv.cvtColor(image, cv.COLOR_BGRA2GRAY)
    _, bright = cv.threshold(gray, 200, 255, cv.THRESH_BINARY)          # Bright parts in white
    contours, _ = cv.findContours(bright, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
    for contour in contours:
        x, y, w, h = cv.boundingRect(contour)
        if x == 0 or y == 0 or x + w == width or y + h == height:
            continue  # Shapes touching the edge of the image are usually the floor
        if cv.contourArea(contour) > 40 and 0.65 < w / h < 1.5:
            return True  # A bright, roughly square shape on a wall
    return False

C/C++

bool seesToken(const unsigned char *image) {
  int height = camera->getHeight(), width = camera->getWidth();
  cv::Mat frame(height, width, CV_8UC4, (void *)image); // BGRA image for OpenCV
  cv::Mat gray, bright;
  cv::cvtColor(frame, gray, cv::COLOR_BGRA2GRAY);
  cv::threshold(gray, bright, 200, 255, cv::THRESH_BINARY); // Bright parts in white
  vector<vector<cv::Point>> contours;
  cv::findContours(bright, contours, cv::RETR_EXTERNAL, cv::CHAIN_APPROX_SIMPLE);
  for (const auto &contour : contours) {
    cv::Rect box = cv::boundingRect(contour);
    if (box.x == 0 || box.y == 0 || box.x + box.width == width || box.y + box.height == height)
      continue; // Shapes touching the edge of the image are usually the floor
    double ratio = (double)box.width / box.height;
    if (cv::contourArea(contour) > 40 && ratio > 0.65 && ratio < 1.5)
      return true; // A bright, roughly square shape on a wall
  }
  return false;
}

This is a simple start: it cannot tell the symbols apart, and it does not see cognitive targets well. The Wall Tokens tutorial shows how to do better.

5. Reporting

To score a token, the robot must stop for more than 1 second, and send its position and the type of the token. Each token can only be scored once, and reporting it again costs 5 points, so the program remembers where it has reported. Communication explains these messages.

Python

def report(tokenType):
    leftWheel.setVelocity(0)  # Stop for more than 1 second
    rightWheel.setVelocity(0)
    delay(1300)
    x = int(gps.getValues()[0] * 100)  # Meters to centimeters
    z = int(gps.getValues()[2] * 100)
    emitter.send(struct.pack("i i c", x, z, tokenType.encode()))
    robot.step(timeStep)
    reported.append((x, z))

def alreadyReported():
    # Reporting the same token again is a misidentification
    x = gps.getValues()[0] * 100
    z = gps.getValues()[2] * 100
    return any(abs(x - rx) < 12 and abs(z - rz) < 12 for rx, rz in reported)  # Closer than 1 tile

C/C++

void report(char tokenType) {
  leftWheel->setVelocity(0); // Stop for more than 1 second
  rightWheel->setVelocity(0);
  delay(1300);
  int x = (int)(gps->getValues()[0] * 100); // Meters to centimeters
  int z = (int)(gps->getValues()[2] * 100);
  char message[9];
  memcpy(&message[0], &x, 4);
  memcpy(&message[4], &z, 4);
  message[8] = tokenType;
  emitter->send(message, 9);
  robot->step(timeStep);
  reported.push_back({x, z});
}

bool alreadyReported() {
  // Reporting the same token again is a misidentification
  double x = gps->getValues()[0] * 100, z = gps->getValues()[2] * 100;
  for (const auto &position : reported)
    if (fabs(x - position[0]) < 12 && fabs(z - position[1]) < 12) // Closer than 1 tile
      return true;
  return false;
}

6. The main loop

Finally, the main loop puts everything together. Each step, it starts at full speed, turns away from walls, turns back from holes, and reports what the camera sees when it is close to a wall in front.

Python

while robot.step(timeStep) != -1:
    speeds[0] = maxVelocity
    speeds[1] = maxVelocity

    if leftDist.getValue() < 0.05:   # Wall on the left: turn right
        turnRight()
    if rightDist.getValue() < 0.05:  # Wall on the right: turn left
        turnLeft()
    if frontDist.getValue() < 0.05:  # Wall in front: spin
        spin()

    if getBrightness() < 80:         # Black floor: turn away from the hole
        spin()
        leftWheel.setVelocity(speeds[0])
        rightWheel.setVelocity(speeds[1])
        delay(600)

    # Close to a wall, and something on it: report it
    if frontDist.getValue() < 0.07 and not alreadyReported() and seesToken(camera.getImage()):
        report("H")  # The type cannot be told yet, so guess "H"

    leftWheel.setVelocity(speeds[0])
    rightWheel.setVelocity(speeds[1])

C/C++

int main() {
  enableSensors();
  leftWheel->setPosition(INFINITY); // Let the wheels turn without limit
  rightWheel->setPosition(INFINITY);

  while (robot->step(timeStep) != -1) {
    speeds[0] = maxVelocity;
    speeds[1] = maxVelocity;

    if (leftDist->getValue() < 0.05) // Wall on the left: turn right
      turnRight();
    if (rightDist->getValue() < 0.05) // Wall on the right: turn left
      turnLeft();
    if (frontDist->getValue() < 0.05) // Wall in front: spin
      spin();

    if (getBrightness() < 80) { // Black floor: turn away from the hole
      spin();
      leftWheel->setVelocity(speeds[0]);
      rightWheel->setVelocity(speeds[1]);
      delay(600);
    }

    // Close to a wall, and something on it: report it
    if (frontDist->getValue() < 0.07 && !alreadyReported() && seesToken(camera->getImage()))
      report('H'); // The type cannot be told yet, so guess 'H'

    leftWheel->setVelocity(speeds[0]);
    rightWheel->setVelocity(speeds[1]);
  }

  delete robot;
  return 0;
}

How well does it do?

This robot wanders: it does not know where it has been, and it guesses the type of every victim. It gets stuck sometimes, and the game moves it back to a checkpoint after 20 seconds. That is fine for a first program.

Try it

• Find the best speeds and distances for your robot.
• Follow the wall on one side: keep the side sensor at a constant distance, and turn when the front sensor sees a wall.
• Change the speed depending on how close the walls are.

Then, learn to score points in Communication.