Skip to main content

Lego Robot and Neural Networks

An overview of using Lego RCX  robots for teaching neural networks present at workshop in 2011.

The video below shows the robot trying out sets of weights for two neurones, until a set of weights are found that enable the robot to go around the circle.

As a part of a set of tools I have found the following useful for teaching the principles of simple neurones.

 Example code:

import josx.platform.rcx.*;

public class annlf{
 public static void main(String[] args)
  int w[][] ={//put weights here};
  int o[]={1,1};
  int s1,s2,res1,res2;
  int sensor1=0,sensor2=0;
  robot_1 tom=new robot_1();
   if (sensor1<42)
   if (sensor2<42)
   if (res1>=0)
   if (res2>=0)
   if ((o[0]==1)&&(o[1]==1))
   if ((o[0]==0)&&(o[1]==0))
   if ((o[0]==1)&&(o[1]==0))
   if ((o[0]==0)&&(o[1]==1))

The example code uses two neurones to produce a line follower. The nice thing about this though is it easy to adapted this for a single neuron or multiple neuron tasks. For more on this some examples can be found here.
The above approaches used the Mindstorms RCX robots but it can equally be done with the newer NXT robots

 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.

Popular posts from this blog

Micro:bit, Servo control with Micropython or blocks

You can control servos (small ones) from a Micro:Bit directly. Following a link from the David Whale (Twitter ) , thank you, took me to a Kitronik blog post,, which has the answer.

The code uses Microsoft Blocks taken from the post, runs the servos 180 degrees and back again, when button A is pressed. It does exactly what it should. I am also using the Tower Pro SG90 servo.
Can it be replicated in Micropython? This is a new mini project, there seems to be little out there yet on how do this but the best so far is this video by PHILG2864:

The closest I have is the following, it is essentially there.
from microbit import *
while True:

Setting the time period to 20ms  pin0.set_analog_period(20)seems by experiment (and used in the video above) to be best value so far. The reason for pin0.write_analog(1)  set to 1 i…

mbots - graphical programming and Arduino

Makeblock ( funded through Kickstarter the development of a new robot - mBot ( with the subtitle "$49 educational robot for each kid". What they came up with is a interesting system that uses their mBlock software, which resembles Scratch but produces code for Arduino, to program a robot with LEDs, light sensors and buzzer integrated on the main board; but also comes with sensors for line-following, ultrasonic sensor and with the version in the kickstarter reward a 16x8 LED matrix.

My impression so far it is really quite intuitive to work with, in the example above the robot:

moves forward;displays 'f' on the LED matrix; turns right;displays 'r' on the LED matrix;repeats until the on-board is pressed to stop the motors. 

What I like most though is seeing the graphical code turned into Arduino code - the potential to see the same thing done into two ways…

4Tronix Bit:Bot Neuron Controlled Edge follower

In thelast post I was playing with 4Tronix'sBit:Bot. In this post I will show the initial experimentation with an artificial neuron controlling the Bit:Bot to follow the edge of a line (it follows the left-hand side of the line).

The neurons (well two separate ones, S1 and S2) are produced using weighted sums - summing the weights x inputs [ right-hand sensor (rs) and left-hand sensor (ls)] plus a bias for each neuron in this case w[0] and w[3].

    net=w[0]+w[1]*rs+w[2]*ls           net2=w[3]+w[4]*rs+w[5]*ls

  If weighted sum >=0 then its output 1 otherwise 0 if net>=0:          s1=1     else:         s1=0
    if net2>=0:         s2=1     else:         s2=0
What actual causes S1 to be either 1 or 0 is all defined by a set of weights w (three for the first neurone, S1,  three for S2).

Converting the outputs of the two neurones S1 and S2 into actions is shown below.