Power plants require periodical inspections to control their state. To ensure a safe operation, parts that could fail before the next inspection are repaired or replaced, since a forced outage due to a failure can cost up to millions of dollars per day. Non-Destructive Testing (NDT) methods are used to detect different defects that could occur, such as cracks, thinning, corrosion or pitting. Some parts are inspected directly in situ, but may be difficult to access; these can require opening access holes or building scaffoldings. Other parts are disassembled and inspected in workshops, when the required inspection tools cannot be moved. In this thesis, we developed innovative miniature mobile systems able to move within these small and complex installations and inspect them. Bringing sensors to difficult-to-access places using climbing robots can reduce the inspection time and costs, because some dismantling or scaffolding can be eliminated. New miniature sensors can help to inspect complex parts without disassembling them, and reduce the inspection costs, as well. To perform such inspections, miniature mobile systems require a high mobility and keen sensing capabilities. The following approach was used to develop these systems. First, different innovative climbing robots are developed. They use magnetic adhesion, as most structures are made of ferromagnetic steel. Then, vision is embedded in some of the robots. Performing visual inspections becomes thus possible, as well as controlling the robots remotely, without viewing them. Finally, non-visual NDT sensors are developed and embedded in some of the robots, allowing them to detect defects that simple vision cannot detect. Achieving the miniaturization of the developed systems requires strong system integration during these three steps. A set of examples for the different steps has been designed, implemented and tested to illustrate this approach. The Tripillars robots, for instance, use caterpillars, and are able t