| dc.description.abstract | Timing is an important aspect when working with visual odometry (VO). The time of processing and data-transfer of image-based measurements are significantly longer than IMU, wheel encoders, and GPS measurements. This introduces an arrival latency, the VO measurements require delay fusion, a subcategory of out-of-sequence measurement (OOSM) fusion. In cooperation with Adigo AS this thesis has focused on the Asterix project, where an agricultural robot uses a downward facing camera for visual-inertial odometry to aid in localization. The main focus in this work is with the delay fusion problem. By approaching the OOSM fusion with a Bayesian framework, the theory chapter presents a generalization
of stochastic cloning. A byproduct of the investigation is an unscented multiple-point smoother capable of defining fixed-points to be smoothed on demand. Simulations and experiments showed that the OOSM fusion methods worked, but model inconsistencies and
inaccuracies in the VO measurements negatively affected the results. | |
