@PhdThesis{duepublico_mods_00074235,
  author = 	{Schl{\"o}gl, Barbara},
  title = 	{Force based inclinometry for navigation in legged robots inspired by the desert ant Cataglyphis spec.},
  year = 	{2021},
  month = 	{May},
  day = 	{10},
  keywords = 	{Bionik; Robotik; Hexapod; Cataglyphis fortis; Laufroboter; Navigation; Odometrie; W{\"u}stenameise; Neuronale Netze},
  abstract = 	{In the Tunisian desert ants of the species Cataglyphis fortis exhibit remarkable navigational abilities. They are able to home even under extreme conditions and rough terrain without the use of external guidance. This makes them ideal candidates for model organisms when developing biomimetic navigation strategies for mobile robots. This approach tackles challenges routinely faced in autonomous mobile robot navigation. The system is robust to external influences, computationally cheap, self-contained and independent of external infrastructure. The broader goal is the development of a generic biomimetic system capable of autonomous locomotion. In particular this thesis addresses the element of detecting the substrate inclination which is crucial information in earth-bound navigation in three-dimensional space. This information will be used in an odometer to determine the ground projection of the travelled distance and thus allow reliable path integration in conjunction with compass information. Based on existing research like behavioural experiments on the desert ants the hypothesis is developed that measuring forces on the legs during movement is the determining factor to gain sufficient information about the substrate inclination. In the course of this thesis it is narrowed down to the measuring of only joint torques to reach that goal. The thesis follows the biomimetic engineering process. The model organism is analysed with respect to morphology, location and distribution of force sensors campaniform sensilla with a scanning electron microscope. The findings are validated via synchrotron X-ray microtomography analysis. The essential leg movement is extracted from digitized high speed videos. The data is used to establish an automated system to translate the motion to joint angles under varying constrictions that can be chosen based on the design of the desired technical application. A one legged prototype is constructed to evaluate the permissible level of abstraction and the suitability of force sensors versus torque sensors. The results are subsequently transferred to a generic commercially available mid range hexapod platform. Manual analysis of joint torque proves the general feasibility of correlating the joint torque measurements with the substrate inclination. To eliminate designing expert systems for each eventuality encountered in an exploratory mission artificial neural networks are employed. Simulated data is used to optimize the topology of a shallow feedforward network sufficient for this task. Loading and unloading processes as well as slanted paths are covered in this approach. Validation is performed on experimental data. The final conclusion of the thesis affirms the initial hypothesis. It is possible to implement a system on a generic mobile robot platform that allows detection of the substrate inclination based solely on idiothetic cues, i.e. joint torques. The proposed system works without installation of additional sensors and is computationally economical through the use of artificial neural networks. Determining the inclination is possible with a temporal resolution of once per step and as such suited for real time capable systems. The presented method has implications on biological research as well as the development of generic biomimetic systems. New hypotheses about the desert ants can be generated and experiments designed with the use of the robotic platform. Furthermore the proposed method ties in seamlessly with existing research of two-dimensional navigation systems or as a redundant system to amend alternative navigation methods.},
  doi = 	{10.17185/duepublico/74235},
  url = 	{https://duepublico2.uni-due.de/receive/duepublico_mods_00074235},
  url = 	{https://doi.org/10.17185/duepublico/74235},
  file = 	{:https://duepublico2.uni-due.de/servlets/MCRFileNodeServlet/duepublico_derivate_00073989/Diss_Schloegl.pdf:PDF},
  language = 	{en}
}