Implementation of multi-algorithm controllers for path determination in mobile robot systems
Recent advancements in control systems, such as the ones used in missile technology in the military or autonomous vehicle development have motivated this study in an attempt to explore various control algorithms and their implementation relevant those applications. Both missile interceptor and autonomous vehicle technology require precise and responsive control system to accurately determine the projectile path of pursuer to strike a moving target or reach a static finish line.The objective of this study is to investigate the performance of several control techniques for a mobile robot to autonomously track and pursue a moving object. Computer model is developed to numerically predict the path taken by the pursuer as it tracks an object moving in regular or random manner. In the computer simulation, the robot's path is calculated using three different techniques: reactive controller, linear estimation, and artificial neural network. Fitness of each method may be determined by evaluating the controller against several factors, such as interception time, steady-state positional error, steady-state time (settling time) and algorithm complexity, listed in decreasing order of importance. A working experimental model is developed to validate the controller selection determined from the computer model simulation. In the experimental setting, the primary inputs to the robot are visual images from cameras. The experiments are carried out with the robot receiving visual inputs from two different perspectives, overhead and frontal vision. Robust image processing technique becomes a topic of significant importance for the system. To manipulate visual images in real-time from raw inputs to comprehensible data, while maintaining fast computational time is a challenge that is addressed in this study. The results from computer simulations show that artificial neural network is a more powerful control algorithm, capable of estimating the object's path more accurately than the other two controllers, resulting in smaller steady-state positional error. The experimental results confirm this conclusion as artificial neural network outperforms the reactive and linear controller by intercepting the object more quickly, i.e. shorter interception time.