The end state of an autonomous system in South Africa's deep mines is a “fait accompli”. The current unacceptable safety records, and the increasing dangers as the mines get deeper, necessitate the removal of miners from the dangerous stope areas. Robotics seems an obvious solution. An autonomous robotic system to inspect the mine ceiling (hanging wall) is being developed at the Center for Mining Innovation as an initial robotic application for South African deep gold mines. A number of the key technologies needed to enable this are discussed. The localization system, the underground alternative to the GPS, is perhaps the single biggest hurdle needed in enabling underground robotics. A low cost, disposable solution for the small area gold stope (30m × 3m) is presented. Machine sensing of both the environment and of humans is critical in a shared working environment. Here we discuss alternatives to the current sensors used above ground for machine perception. In the deep gold mines the geothermal heat result in hot walls and thermal imaging becomes an option for structural imaging. The combination of temperature with a 3D data enables the determination of a risk measure, indicating potential danger areas. Potential methods of representing the risk data for the miner to interpret are discussed. Finally, the thermal camera in conjunction with a distance sensor is used to identify and track pedestrians in order to predict potential collisions.