Neural network based control of a four rotor helicopter

This thesis presents the design and development of an intelligent controller based on neural networks for a hoverable flying robot to be capable of achieving a stable hovering position in three axes. Control of the fourth axis, height, and the ability to be able to autonomously navigate through a predefined path was designated for a future phase.; This work is different from most autonomous flying robots as it focuses on a four-propeller configuration. This is a very rare helicopter design because of its inherent instability and it is believed that an autonomous robot of this configuration has not yet been successfully developed. In addition, this project uses fixed pitch propellers instead of variable pitch rotors, resulting in a greatly reduced cost and mechanical complexity. The downside to this design is that this introduces significant additional challenges in the control. Relative stability was achieved in three axes and all the supporting modules were successfully designed and implemented. However, significant challenges were encountered including the complexities of creating a Neural Networks Controller (NNC) to work in real-time in a slow microcontroller as well as to develop the training process.