I wrote this paper for a RSSC presentation of how basic neural networks work and how they can be implemented in robotics. The network is of a simple structure and follows some of the experiments described by Ulrich Nehmzow on a robot called Alder. The robot that I used was Sparky. Developing the software was challenging because the Basic Stamp does not use floating point so everything had to be converted to use integer math.
The code can be found here: nueral_net2.bs2.txt
Here is a video of the robot in progress. The robot moves towards the wall and tires a move. When the move fails (the sensors are still on) the robot tries a another move. Since this move worked, next time the robot encounters the same obstacle it "knows" what to do.
