Description

Robots are physical systems with varying degrees of autonomy that operate in different and dynamic physical environments. Their use in our daily lives is increasing, as it is appealing for tasks that can be referred to as the four Ds —too Dangerous, too Dull, too Dirty, and too Difficult— to be done by humans.
Nevertheless, robotic systems are prone to different types of faults, which have the potential to affect the efficiency and the safety of the robot and/or its surroundings. For these reasons, FDD (Fault Detection and Diagnosis) techniques are nowadays essential in robotics, with the aim of facilitating the system recovery.
Based on such considerations, this thesis addresses the problem of supervision of a humanoid robot, specifically focusing on its head. With this scope in mind, the robotic system has been modelled and controlled by means of a linear parameter varying (LPV) feedback controller.
Hence, a fault detection and isolation scheme has been implemented using the LPV approach. Such a method has been selected as the one to be followed as it encompasses the performance requirements a humanoid robot implies: it has to detect faults quickly, online and with a low computational burden, according to expectations autonomously generated.
Later, a fault tolerant scheme has been designed to compensate the faulty effect, once the fault is detected and isolated. Lastly, all the above-mentioned schemes have been tested in simulation.

https://upcommons.upc.edu/handle/2117/120457

The work is under the scope of the following projects:

• MdM: Unit of Excellence María de Maeztu (web)
• HuMoUR: Markerless 3D human motion understanding for adaptive robot behavior (web)