Learning data-derived vehicle motion models for use in localisation and mapping
Year: 2018
Type: dissertation
Abstract: Various solutions to the Simultaneous Localisation and Mapping (SLAM) problem have been proposed \nover the last 20 years. In particular, extending the fundamental solution of the SLAM problem has \nattracted a great deal of attention. Most extensions address shortcomings such as data association, \ncomputational complexity and improving predictions of a vehicle’s state. However, nearly all SLAM \nimplementations still depend on analytical models to provide estimates for state transitions. \nLearning data-derived non-analytical models for use during localisation and mapping provides an \nalternative that could significantly improve estimates and increase the flexibility of models. A methodology \nto learn motion models without knowledge of the higher-order dynamics is therefore proposed \nusing tapped delay-line neural networks (TDL-NN). Incorporating the learned Nth-order Markov \nmodel into a recursive Bayesian estimator requires that the learned model be assumed independent of \nthe transitional model, forming a black box estimator. Both real-world and simulated training data \nwere evaluated, along with changes to the input data’s format, to determine the best vehicle motion \npredictor. Furthermore, an evaluation methodology is defined to asses how well the models could learn each \nmotion type. A comprehensive analysis of the one-forward prediction using various statistical measures \nwas used to determine the most appropriate metric. The methodology evaluated the predictions at \ndifferent levels of depth, providing supplementary information on the type of motions that are learnable. \nOutcomes of the experiments revealed that inherently learning a vehicle’s dynamics cannot be achieved \nusing TDL-NNs. Currently the best that such an approach can learn is the delta between the vehicle’s \nstates. Consequently, modifications are required to the learning algorithms as well as the input data’s \nformat that will force the strategies to learn the higher-order dynamics.
Author: Kobie Wood Fick
Institution: University of Pretoria
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2907995004
