Analysis and Application of Computer Vision Principles towards Self-replicating Robots

Abstract

Robots are used to assist or replace humans in dangerous and repetitive tasks like space exploration. Over the last few years, researchers have studied the feasibility of self-replicating robots. Replicating a robot with high complexity has proven to be difficult. This master thesis presents a complex self-replicating robot and its vision system. The necessary physical and logical components are analyzed.A modular robot prototype is designed and developed using the LEGO MINDSTORMS EV3 set. The robot can be replicated by assembling its modules. The robot can move in an unstructured environment, where resource modules are spread randomly. The robot uses a gripper to gather the modules in a location for assembling. To recognize different objects, the robot is equipped with a camera. A remote software application is used to recognize the required modules and control the robot. OpenCV library is used to acquire, analyze, process and recognize objects in the video frames. The object recognition algorithm is color-based since the resource modules are colored LEGO bricks. The control algorithm sends commands to the robot based on the object’s location. The complexity and degree of self-replication are analytically calculated. The self-replication degree of the robot prototype presented in this thesis is higher than in past research. This shows the progress made by this research towards self-replicating robots consisting of a great number of low complexity modules, able to function in an unstructured environment. The software application demonstrated high performance and accuracy. Tests identified a direct relationship between the processing time and number of object groups. The accuracy decreases in continuously changing light conditions.