The use of autonomous robots in industrial applications is growing in popularity and possesses the following advantages: cost effectiveness, job efficiency and safety aspects. Despite the advantages, the major drawback to using autonomous robots is the cost involved to acquire such robots. It is the aim of GMSA to develop a low cost AGV capable of performing material handling in an industrial environment. Collective autonomous robots are often used to perform tasks, that is, more than one working together to achieve a common goal. The intelligent controller, responsible for establishing coordination between the individual robots, plays a key role in managing the tasks of each robot to achieve the common goal. This dissertation addresses the development of an AGV capable of such functionality. Key research areas include: the development of an autonomous coupling system, integration of key safety devices and the development of an intelligent control strategy that can be used to govern the operation of multiple AGVs in an area.

Mining is one of the main economic sectors in South Africa. Mining activity contains hazards such collapsing of structures, presence of dangerous gases, accidental explosions and fires. Even though most of these hazards are identified and minimized sometimes accidents occur. These accidents lead to human injuries, direct fatalities and fatalities resulting from delays in victims getting medical attention as a result of delays in search and rescue missions. The rescue missions in underground mines present challenges where rescuers are not sure which locations are victims in, what the area conditions like in the rescue path. A quad rotor unmanned aerial vehicle (UAV) for search and rescue missions is presented. The UAV is controlled from a remote location over Wi-Fi. The communication allows data relay to the ground control station. The communication system is tested on the university’s Wi-Fi network. The UAV also contains a vision system that contains a human detection algorithm to give an indication of human presence to rescuers. The human detection system is based on Haar- Cascade classifiers. The model developed was found to have a false alarm rate of 5×10-3% after training. The model was further tested on streaming data and the overall average positive human detection was found to be 97 %. In the same tests overall false average detection was found to be 2.5 %. The video feed is streamed from the UAV to the ground station (GS) and the flight control instructions are sent to the UAV from the GS via Wi-Fi. / Thesis (M.Sc.Eng)-University of KwaZulu-Natal, Durban, 2013.