Sensor fusion, mobile autonomy, and long-term autonomy.
Papers and presentations
- Automated Topological Mapping for Agricultural Robots – Presented by co-author Willow Mandil at UKRAS2020.
- Lincoln Conference on Intelligent Robots and Systems 2020 – Semantic Segmentation of Plant Leaves from 3D Point Clouds using Deep Learning (read more here).
- AgriFoRwArdS CDT Annual Conference 2021.
- Member of the AgirFoRwArdS CDT Drink Outside the Box Organisation Committee.
- Member of the AgriFoRwArdS CDT Advisory Board.
- Member of the AgriFoRwArdS CDT Equality, Diversity and Inclusion Panel.
My name is Karoline, I am from Mönchengladbach in Germany, and I joined the CDT in September 2019. I have a background in neuroscience and robotics and I am especially interested in the research areas of sensor fusion, mobile autonomy, and long-term autonomy. In my spare time, I enjoy cooking, sewing (I also like to add pockets to all my clothes 😄 ), playing music, and video games.
I chose to join this CDT because the development of robots for agricultural applications is an interesting real-world challenge that could have a great positive impact on the global food chain. So far I have especially enjoyed the many associated training opportunities and to study alongside a brilliant cohort of fellow PhD students with different backgrounds.
During my first year in Lincoln, I also spent a lot of time exploring the city and its sights, and some of my favourite places are all the little book shops and cafes in the historic quarter. In 2020 I will begin the full-time research phase at the University of Lincoln. There are lots of researchers in Lincoln who I want to work with and I look forward to some hands-on work at the university’s farm facilities at Riseholme Campus.
Semantic Segmentation of Plant Leaves from 3D Point Clouds using Deep Learning
In this project we aim to address the problem of semantically segmenting plant leaves from the background and other plant organs in three-dimensional point clouds of individual plants captured by RGBD sensors. Previous work utilises explicit prior knowledge about the expected plant morphology and sensor set-up, as well as manually tuned parameters to achieve this segmentation. Here we propose to train a supervised machine learning algorithm to predict the segmentation output directly from point cloud data and minimise the necessary user input.
4D Scene Analysis for Autonomous Operation of Mobile Robots on Farms
This research addresses challenges in 4D scene analysis for autonomous operation of mobile robots on farms. The deployment of agricultural robots will increase sustainability and support precision farming operations tuned to needs of individual plants. This research will enable robots to maintain precise 3D representations of uncertain and highly variable farm environments together with their semantics in order to safely traverse the farm autonomously, as well as to reconstruct structural representations of the crop. Additionally, registering these representations over time would allow for sustained crop monitoring and provide insight into spatio-temporal dynamics and interactions between the plants and environmental factors.