Neural Net Controller for a Snake Robot

Abstract

When we re talking about the field of robotics, one of the key components is the controller for the robots. The focus has recently begun shifting away from the traditional wheeled robots, into biologically inspired robots. Research is being done on walker robots with different numbers of legs, and also serpentile robots are being worked on. The previous and reasonably well-established robot controllers for the wheeled robots is generally not applicable for the new robots, since it is now a much larger focus on the motorics and how the robot must manipulate its limbs to be able to move. This opens up a new and more advanced area of research, where both navigation and movement have to be taken into consideration when making the controllers.This project will attempt to make a controller for serpentile robots using an Evolutionary Algorithm to evolve the network on its own, since this has shown promise in previous projects in the field. We also introduce a simple environment into the system to see how the system is able to handle and adjust to changes in the terrain. Primarily we will be looking at changes in elevation, but impassable terrain is also possible using the scheme we implemented. The snake robot is mainly just trying to move as far as possible in any direction, but we will also try giving it specific target destinations.Even if there was some trouble with the simulator, the project did produce positive results. We will also discuss possible ways to expand on this project, or new projects in the future.