In this lab, you will design a protocol, and run a series of experiments to test how sensor noise affects a robot's behavior. To get a copy of this notebook, open a terminal and type:
cp /home/meeden/public/SensorNoiseExperiments.ipynb s3p
This will copy the notebook from my directory to your s3p directory. Then type:
cd s3p
source /usr/swat/bin/s3penv
jupyter notebook
Then you can open this notebook and begin work.
from jyro.simulator import *
from math import pi
from random import random
from time import time
The world consists of a maze of twisting cooridors with a light at the end. The robot's goal is to reach the light as fast as possible.
def make_maze_world(physics):
physics.addBox(0, 0, 5, 5, fill="backgroundgreen", wallcolor="gray")
physics.addBox(1.5, 2, 1.7, 3.5, fill="blue", wallcolor="blue")
physics.addBox(3.0, 5, 3.2, 3.5, fill="blue", wallcolor="blue")
physics.addBox(1.5, 2, 3.0, 1.8, fill="blue", wallcolor="blue")
physics.addBox(3.0, 2, 3.2, 0, fill="blue", wallcolor="blue")
physics.addBox(3.0, 3.5, 5.0, 3.7, fill="blue", wallcolor="blue")
physics.addLight(4, 1, 1.0) #paramters are x, y, brightness
The robot is equipped with a camera, sonars, and light sensors.
def make_robot():
robot = Pioneer("Pioneer", 2, 0.5, 0) #paremeters are name, x, y, heading (in radians)
robot.addDevice(Camera())
robot.addDevice(Pioneer16Sonars())
light_sensors = PioneerFrontLightSensors(3) #parameter defines max range in meters
robot.addDevice(light_sensors)
return robot
robot = make_robot()
vsim = VSimulator(robot, make_maze_world)
One of the biggest problems with using simulated robots rather than real robots, is that the real world is much more complex than a simulated one. In addition, real sensors are often noisy rather than clean. In other words, they are often inconsistent, giving random data at times.
One approach to deal with this discrepancy between simulations and the real world is to explicitly add random noise to the simulator. The function noisySonars, defined below, has two parameters: the robot and the percent noise to add to the robot's sonar values. To add no noise, use 0 as the percentNoise. To add 10% noise, use 0.1 as the percentNoise. To add 20% noise, use 0.2 as the percentNoise. And so on.
def noisySonars(robot, percentNoise):
sonars = robot["sonar"].getData()
noisy_sonars = sonars[:]
for i in range(len(sonars)):
noise = percentNoise * sonars[i] * random()
if random() > 0.5:
noise *= -1
noisy_sonars[i]+= noise
return noisy_sonars
The brain below is designed to traverse a maze, timing how long it takes to reach a light. In your experiments you will manipulate the amount of noise applied to the sonar sensors and measure how this affects the time it takes the robot to reach the light.
This brain has many different cases that it checks when deciding how to move the robot. These cases are always checked in order. The first case that is true will be executed, and all remaining cases will be skipped. The overarching idea is that when there is an obstacle in front, the brain chooses to turn towards the side where there is more open space. And once it starts a turn in one direction, it continues to turn in that same direction until the front is no longer blocked.
global leftTurn, rightTurn, startTime, endTime, begin, done
begin = True # Is set to False as soon as the startTime is set
done = False # Is set to True once the experiment is complete
leftTurn = False # Is set to True when initiating a left turn, and back to False when completed
rightTurn = False # Is set to True when initiating a right turn, and back to False when completed
def avoidToLightBrain(robot):
global leftTurn, rightTurn, startTime, endTime, begin, done
light = robot["light"].getData()
sonars = noisySonars(robot, 0.0) # <-------- MANIPULATE THE AMOUNT OF NOISE HERE
front = min(sonars[2:6])
left = min(sonars[15], sonars[0])
right = min(sonars[7:9])
if begin:
# at the very beginning set the starting time of the experiment
startTime = time()
begin = False
elif done:
# when done, print the elapsed time
print("Time to find light", endTime-startTime, "seconds")
elif sum(light) > 1.5:
# stop when light is found and set done to be True
robot.move(0,0)
endTime = time()
done = True
elif leftTurn and front > 0.5:
# when front becomes clear while turning left, end turn
leftTurn = False
robot.move(1.0, 0)
elif rightTurn and front > 0.5:
# when front becomes clear while turning right, end turn
rightTurn = False
robot.move(1.0, 0)
elif leftTurn:
# continue turning left
robot.move(-0.05, 0.75)
elif rightTurn:
# continue turning right
robot.move(-0.05, -0.75)
elif front < 0.5 and left > right:
# when front is blocked and left side is more open, start left turn
leftTurn = True
robot.move(-0.15, 0.75)
elif front < 0.5 and right > left:
# when front is blocked and right side is more open, start right turn
rightTurn = True
robot.move(-0.15, -0.75)
else:
# otherwise go straight
robot.move(1.0, 0)
robot.brain = avoidToLightBrain
Give a detailed protocol for your experiments here. Be sure to run multiple experiments at each level of noise that you decide to test.
Present the results of your experiments here. Be sure to calculate the average time needed to reach the light for each level of noise tested. Use a markdown table to clearly present the results.
Summarize your results here. Based on your results, how did the amount of noise affect the robot's behavior?