When: Thurs 6th of Aug, 1p AEST
Where: This seminar will be presented online, RSVP here.
Speaker: Jennifer Wakulicz
Title: Estimating unmodelled motion dynamics for robust crop flower thinning
Abstract: Flower thinning is an important practise in orchard management, allowing farmers to pre-determine fruit setting and quality come harvest. Automation of this task requires accurate estimation of flower’s positions — a difficult prospect considering the lightweight nature of crop flowers and their vulnerability to environmental forces such as wind. Each branch in an orchard will respond individually to a gust of wind, and each gust of wind will be of varying direction and strength, making it cumbersome to model the effect of wind on a particular target flower. In this talk I will outline my work towards planning a sensor trajectory that enables the most accurate estimation of flower position under such circumstances without modelling wind dynamics. My approach is to instead treat wind as an unknown input in the flower’s motion model, using an unknown input filter to estimate the flower position. Feeding this estimate into a known active perception algorithm of tuneable optimality/computational complexity gives promising results, with known sub-optimality bounds for the chosen sensor trajectory.
Bio: Jen graduated from the University of Sydney with an undergraduate degree in Mathematics & Physics in 2018 and is now a PhD student at the ACFR, working on active information acquisition algorithms for orchard management.