Student’s projects at Mobots/Learnge

The winter period can be considered the most critical time of the year for a bee colony, where approximately 50% of colonies don’t survive. In the research project HIVEOPOLIS, we are developing a smart hive including detailed sensing that aims to better understand honeybees and support them. In recent winter seasons, a robotic device was used to collect data and interact with specific collective behaviours of groups of thousands of bees.

We aim to better understand how this robotic system can interact with an animal society, and predict how animals will behave under specific environmental conditions. Thus, we propose a project focused on the development of agent-based models that will take into account specific behaviors of winter bees (e.g., movement over thermal gradients, heat production, clustering/aggregation, etc.) and some physical properties of the hive (e.g., heat propagation through honeycombs). Examples of target predictions include temperature profiles, collective metabolism, collective location and consumption of the food supplies. The model should also cater for the sensing and control of the robotic system, thereby bringing together the physical and control elements of the artificial system with the natural system.

In order to extract behavioral parameters and compare simulation results and model accuracy, the student will have access to data collected from real hives from different instrument types (e.g., cameras and temperature sensors).

The HIVEOPOLIS project is developing several robotic systems to monitor and interact with honeybees inside their hives, aiming to better understand and support these fascinating superorganisms. Various prior works have shown that vibrations are important in bee-bee communication, as well as being useful for colony state identification including important health indicators. We have incorporated vibration sensors into honeybee hives to gather vibration data, while simultaneously recording images to see what the bee colony is doing.

This semester project aims to develop a signal processing pipeline to apply to vibrational data from inside beehives. To provide “ground truth” information on what behaviours are generating given vibration patterns, image-based data is available, and will require some level of manual or automated processing.  The goal of the signal processing is to identify signatures corresponding to specific individual and group-level bee behaviours.

• Become familiar with literature on vibrations known in honeybees, as well as typical analysis methods
• Implement time/frequency analysis on 1,2, and 3-channel data; and on multi-sensor data
• Implement signal-noise reduction via multi-sensor data
• Extract ground-truth behavioural state information from video data (likely a combination of computer vision and manual labelling)
• Compare vibration-derived outputs to ground-truth outputs using appropriate metrics

We can support either one or two students working on this project, e.g. by partitioning the target behaviours that each student aims to investigate.

• Taken classes on digital signal processing, pattern recognition,  statistical inference, applied data analysis
• Interest in honeybees, ecosystems, or agriculture