idea is to create obstacles try to avoid obstacles as if you are the robot Neuronal network will use your way to learn how to do the same job backpropagation is used as learning algorithm ones ANN learn, you can see it on the field the software is made in away to explore the effect of accuracy on backpropagation you can any time switch configure the parameter of the arithmetic to use fixed point or floating point and configure how many bit to be in use a hardware FPGA - VHDL implementation will use the tool to determine the minimum precision needed to reduce area overhead : and the Robot avoider is an example to validate the hardware.