Andrew Straw Seminar

Bees are remarkable visual learners. After just a few learning flights, they are capable, repeatedly, of successfully traversing large distances from the nest and returning home with food. Despite these facts being well known for decades, it is still not clear how bees accomplish such feats. This bee learning is much more efficient than state-of-the-art AI training in robotics. We have recently developed a multicopter drone that can follow bees tagged with a small reflector to record videos and high-resolution trajectory data. We intend to use data from such experiments on bees in open, outdoor settings to inform our modeling efforts aimed at making neurobiologically plausible models of navigation. For example, we are interested to understand whether bees form a “cognitive map” of their environment and what computational and neurobiological mechanisms are employed to navigate. Towards this end, we have developed a reinforcement learning agent based on biologically plausible brain structures and neurophysiological responses, and these models are capable of performing simple navigational tasks. Future work will integrate the experimental and theoretical work with an aim of better understanding the neural basis for bee navigation at landscape scale.