Autonomous navigation using neural networks

Semi-autonomous to fully autonomous robots rely on some form of data collection to operate in their environment. This has traditionally been accomplished using sonar or infra-red sensors to measure the robot's proximity to nearby objects. These sensors provide information to the robot so that the software controlling the robot can exhibit some degree of autonomy. These systems commonly use deterministic algorithms that employ rules that attempt to cover any eventuality. This paper discusses an alternative method to this rule based approach. A feed forward neural network was trained to exhibit the same behaviors as a simple rule based algorithm as a first step to a more sophisticated approach that will be able to draw up more complex rule sets, human teaching , and run-time learning that allows the robot to build on past experiences.