Robot control by a neuron.

This year the Computing team has been fortunate enough to host three Nuffield Research Placement students (https://www.nuffieldresearchplacements.org/) all working with Dr Scott Turner.

Michael Welsh

Michael has been working on using a micro:bit based bitbot from 4tronix to produce a potential teaching tool; an example of artificial neurons used control a robot. The aim is for this tool to be used with 3rd-year Undergraduates, as part of a module on Artificial Intelligence.

Michael's solution was to use the computer to run and train a single neuron; then for the robot to send values from the line sensors back to the program running on a Computer and receive control signals. 

Sounds easy? No really, in the end, the software running on the computer had to also send and receive the data through a microbit (via USB) and then use radio to communicate with the bit:bot robot. All the various developed parts of the solution were implemented in Python by Michael.

Example of the code.

import serial
from math import fabs
import random
import sys
import glob


#TO DO: GUI, implement manual mode. Make manual mode a button?? When in automatic, weight boxes are greyed out. Otherwise, they are able to be typed into.

def forward(n):
    microbit.write(('f'+str(n)).encode('utf-8'))
def backward(n):
    microbit.write(('b'+str(n)).encode('utf-8'))
def turnR(n):
    microbit.write(('r'+str(n)).encode('utf-8'))
def turnL(n):
    microbit.write(('l'+str(n)).encode('utf-8'))
def end():
    microbit.write('e'.encode('utf-8'))


def serial_ports():
    """ Lists serial port names

        :raises EnvironmentError:
            On unsupported or unknown platforms
        :returns:
            A list of the serial ports available on the system
    """
    if sys.platform.startswith('win'):
        ports = ['COM%s' % (i + 1) for i in range(256)]
    elif sys.platform.startswith('linux') or sys.platform.startswith('cygwin'):
        # this excludes your current terminal "/dev/tty"
        ports = glob.glob('/dev/tty[A-Za-z]*')
    elif sys.platform.startswith('darwin'):
        ports = glob.glob('/dev/cu.*')
    else:
        raise EnvironmentError('Unsupported platform')

    result = []
    for port in ports:
        try:
            s = serial.Serial(port)
            s.close()
            if '/dev/cu' in port:
                if '/dev/cu.usbmodem' in port:
                    result.append(port)
            else:
                result.append(port)
        except (OSError, serial.SerialException):
            pass
    return result


class neuron:

    def __init__(self,ins):

        self.inputs = [1]

        self.weights = []

        for i in range(ins+1):
            self.weights.append(random.uniform(-1,1))
        #print(self.weights)

        self.LC = 0.005
        
    def get_output(self,inputs):
        self.inputs = [1]+inputs

        WSum = 0

        for i in range(len(self.inputs)):
            WSum += int(self.inputs[i]) * self.weights[i]

        sums = sorted([fabs(WSum), fabs(WSum - (1/3)), fabs(WSum-(2/3)), fabs(WSum-1)])
        #print(sums)
        if sums[0] == fabs(WSum): #return 0 - move backwards
            backward(5)
            return 'B',WSum
        elif sums[0] == fabs(WSum - (1/3)):
            turnR(5)
            return 'R', WSum
        elif sums[0] == fabs(WSum - (2/3)):
            turnL(5)
            return 'L', WSum
        else:
            forward(20)
            return 'F', WSum

    def train(self,inputs,desired):

        result = self.get_output(inputs)[1]
        print(result)
        error = desired - result
        #print(error)

        for w in range(len(self.weights)):

            change = self.LC * int(self.inputs[w]) * error

            print('change in weight ' + str(w) + ': ' + str(change))

            self.weights[w] += change


for i in serial_ports():
    try:
        microbit = serial.Serial(port=i, baudrate=115200)
        break
    except:
        pass

microbit.setDTR(1)
microbit.close()
microbit.open()

control = neuron(2)

mode = input('Automatic or Manual? (A/M)\n')
while mode.lower() not in ['a','m']:
    mode = input('Automatic or Manual? (A/M)\n')
if mode == 'm':
    control.weights[0],control.weights[1],control.weights[2] = int(input('Enter weight 0/bias: ')),int(input('Enter weight 1: ')), int(input('Enter weight 2: '))
while True:
    #microbit.write(input('input: ').encode('utf-8'))
    out = microbit.read(6)
    microbit.flush()
    try:
        out = out.decode()
    except:
        pass
    for i in range(len(out)):
        if str(out[i]) in ['0','1']:
            if i == 0:
                if out[1] == ']':
                    num2 = out[i]
                elif out[1] == ',':
                    num1 = out[i]
            elif i == 5:
                if out[4] == '[':
                    num1 = out[i]
                elif out[4] == ' ':
                    num2 = out[i]
            else:
                if out[i-1] == '[':
                    num1 = out[i]
                elif out[i-1] == ' ':
                    num2 = out[i]
                elif out[i+1] == ']':
                    num2 = out[i]
                elif out[i+1] == ',':
                    num1 = out[i]
        
        #print(out)
    try:
        sleft=int(num1)
        sright=int(num2)
        print([sleft, sright])
        if mode == 'a':
            if [sleft, sright] == [0, 0]:
                control.train([sleft,sright],0)
            elif [sleft, sright] == [0, 1]:
                control.train([sleft,sright], 1/3)
            elif [sleft,sright] == [1, 0]:
                control.train([sleft,sright], 2/3)
            else:
                control.train([sleft,sright], 1)
        else:
            if [sleft, sright] == [0, 0]:
                control.get_output([sleft,sright])
            elif [sleft, sright] == [0, 1]:
                control.get_output([sleft,sright])
            elif [sleft,sright] == [1, 0]:
                control.get_output([sleft,sright])
            else:
                control.get_output([sleft,sright])
    except:
        pass

All opinions in this blog are the Author's and should not in any way be seen as reflecting the views of any organisation the Author has any association with. Twitter @scottturneruon

Robots behaving...

Reblogged from: http://scott-ltattempts.blogspot.co.uk/2011/05/robot-behaviours.html


Behaviour based robots was used in the teaching as way of getting the students to think out AI a little deeper and in particular Do we need Human Level intelligence? or rather Do we always need to aim for Human Level Intelligence?

Lego Mindstorms robot are a good vehicle for students to start trying out idea around behaviour-based robotics. They are inexpensive, programmable and with the LeJOS software installed on them; have behaviours built into the programming which is done in Java.

A good example to use comes from Bagnall's book (B Bagnall (2002) Core Lego Mindstorms: 

Programming the RCX in Java , ISBN: 978-0130093646)

code 1: HitWall

//Taken from Bagnall (2002)
import josx.robotics.*;
import josx.platform.rcx.*;
public class HitWall implements Behavior
{
public boolean takeControl()
{
return Sensor.S2.readBooleanValue();
}
public void suppress()
{
Motor.A.stop();
Motor.C.stop();
}
public void action()
{
Motor.A.backward();
Motor.C.backward();
try{Thread.sleep(1000);}catch(Exception e){}
Motor.A.stop();
try{Thread.sleep(300);}catch(Exception e){}
Motor.C.stop();
}
}

Code 2: DriveForward

//Taken from Bagnall (2002)

import josx.robotics.*;
import josx.platform.rcx.*;
public class DriveForward implements Behavior
{
public boolean takeControl()
{
return true;
}
public void suppress()
{
Motor.A.stop();
Motor.C.stop();
}
public void action()
{
Motor.A.forward();
Motor.C.forward();
}
}



Code 3: Bumper Car

import josx.robotics.*;
public class BumperCar
{
public static void main(String [] args)
{
Behavior b1=new DriveForward();
Behavior b2=new HitWall();
Behavior [] bArray ={b1,b2};
Arbitrator arby=new Arbitrator(bArray);
arby.start();
}
}

All opinions in this blog are the Author's and should not in any way be seen as reflecting the views of any organisation the Author has any association with.

Aldebaran NAO 'Red' in Teaching

Photo by John Sinclair

I had my first opportunity today to try an Aldebaran NAO robot as a teaching tool in an AI class today. The session was an end of term activity around summarising what we did in the AI class so far and questions. 

A question came up around AI and it's impact on society. Perfect opportunity to bring in a social robot - especially as a precursor for when we include a session on social robotics next term.


All opinions in this blog are the Author's and should not in any way be seen as reflecting the views of any organisation the Author has any association with.