Mikschi, M. (2023). Identification of the driving dynamics of a skid-steered mobile robot based on geodetic measurements [Diploma Thesis, Technische Universität Wien]. reposiTUm. https://doi.org/10.34726/hss.2023.112060
The driving dynamics of skid-steered vehicles are difficult to model due to their inherent need for loss of traction for curvilinear motion, which leads to complex wheel-ground interactions. However, such vehicles represent well suited platforms for automated robots with their robust, cost-effective, low maintenance construction and their great off-road performance. One usecase of them is mobile mapping. Such systems can greatly benefit from precise driving dynamic models for pose estimation for both georeferencing measurements and navigation as well as for system control using model predictive control methods.System identification is a field of applied mathematics for estimating models of dynamical systems based on measured input and output data of the system. The SINDY (Sparse Identification of Nonlinear DYnamics) algorithm is a method that utilizes sparse regression to identify interpretable, parsimonious models in state-space representation, that balance model performance with complexity.In this thesis the suitability of the SINDY algorithm to conduct system dentification for the driving dynamics of the Clearpath Husky A200 based on geodetic measurements was ascertained. The Husky A200 is a medium sized robot for research and prototyping and represents an example of skid-steered unmanned ground vehicle (UGV) well suited for tasks such as mobile mapping. A measurement setup around two laser trackers for collecting the necessary data was created, addressing the challenges of time synchronisation of the different system components and maintaining an uninterrupted line of sight between the laser trackers and their target prism during driving operations.A preprocessing pipeline to calculate the system identification input data was established, accomplishing time synchronisation, pose calculation and interpolation as well as state vector calculation and transformation. The system identification was successfully conducted, utilizing the integral notation of SINDY and employing an extensive hyperparameter tuning. The point position estimation uncertainty of the best performing model was 14 cm after a 5 second integration period, with the heading estimation uncertainty being 4.7◦. These results demonstrate the suitability of identified systems for example for certain applications of state estimation. Potential shortcomings and areas for improvements of the presented measurement setup and methodology were identified and discussed.