Postfiltering of honeybee queen marker-based tracking data

Jan Blaha
PhD student
Czech Technical University in Prague, Czechia
Lab webpage:

With drastic changes to the natural environments happening in the world today, the scientific community has started to develop techniques and technologies to monitor and influence living ecosystems, called ecosystem hacking”. To be able to develop such technologies, it is necessary to have methods for processing real-world data, which are known to the robotic community as particularly problematic.
In our projectRoboRoyale”, we aim to develop a biohybrid system influencing the ecosystems by modulating the behaviour of one of the ecosystem’s keystone species, the honeybees, through interaction with the honeybee queen. During our work on semi-laboratory experiments, we have collected data from automatic queen detection and tracking that are high in noise and misdetections, mostly due to natural clutter in the environment. We show that training a simple CNN classifier can dramatically improve the data obtained; in particular, our method can classify queen detections from detections of the surrounding clutter.
Our method achieved good performance with only a very little annotation effort, allowing us to filter out millions of bad datapoints with little human intervention. We share our experience as the in-the-wild experimentation outside of laboratory conditions will be crucial for merging cybernetic technologies with living ecosystems. Such experiments will necessarily bring more and more imperfections in the experimental setup the researchers will have to deal with.

Currently working towards a PhD in informatics at the Czech Technical University, Jan is a data scientist with a great interest in statistics and stochastic modelling. As the domain of his dissertation, he has chosen the field of biohybrid systems modelling and chronorobotics, i.e. the study of spatio-temporal representations for robots.