Improving robotic active perception in agriculture using online self-supervised learning

In modern greenhouses, many important tasks, such as harvesting, pruning, and monitoring plant growth, still depend on manual work. Robots could help, but plants are complex: leaves and stems often hide the fruits, flowers, or other parts the robot needs to see. My PhD research studies how a robot can actively move its camera around a plant to collect better information, instead of relying on one fixed view. The robot learns to choose the next useful viewpoint by itself, using the data it collects during the task. This makes the system faster, more flexible, and better able to deal with different plants and changing greenhouse conditions. The research contributes to more reliable robotic perception in agriculture and brings robots a step closer to practical use in labour-intensive greenhouse work.