Our Contributions

Modular Robot Platform

Our aim was to develop a modular wheeled robot platform (WMR) that enables to easily build robots of almost arbitrary shape and size with a variety of different wheel alignments. In further consequence, this platform should serve as a testbed for model-based control strategies, as well as diagnosis and reconfiguration schemes developed at the institute. These considerations in mind, we came up with the idea of building the robots from hexagonal modules.


On-line kinematics reasoning

Our aim was to devise a kinematics reasoning concept that captures robot drives of almost arbitrary geometry and functionality. For this purpose, we developed a reasoning scheme that analyzes the mobility capabilities of a robot drive at its current mode of operation/failure. Based on this result the (inverse) kinematics for on-line low-level control of the actuators (steering and rotational actuators for the individual wheels) can be deduced. We were able to implement this powerful reasoning scheme on a general purpose real-time system (National Instruments CompactRIO) and thus integrated this functionality within the low-level control loop of our robot drive. In that way we obtained a controller that automatically handles faults and configuration changes within the drive.


Hybrid Diagnosis

Our aim was to develop a generally applicable hybrid diagnosis/estimation scheme with a focus on fast detection and identification of mode changes as well as on run-time performance. A close cooperation with partners at LAAS-CNRS in Toulouse, France (https://www.laas.fr/public/en/disco), resulted in a successful approach that combines our previous work on a dynamic-filter based hybrid estimation (hME) with analytic/algebraic methods developed for discrete event systems (DES) at LAAS (HyDiag). We devised a synergetic hybrid estimation procedure that carefully applies both estimation schemes. This results in an improved performance in terms of estimation quality (continuous state estimation and correct mode identification) and computational effort.