Spatio-temporal Memory for Navigation in a Mushroom Body Model

Insects, despite relatively small brains, can perform complex navigation tasks such as memorising a visual route. The exact format of visual memory encoded by neural systems during route learning and following is still unclear. Here we propose that interconnections between Kenyon cells in the Mushroom Body (MB) could encode spatio-temporal memory of visual motion experienced when moving along a route. In our implementation, visual motion is sensed using an event-based camera mounted on a robot, and learned by a biologically constrained spiking neural network model, based on simplified MB architecture and using modified leaky integrate-and-fire neurons. In contrast to previous image-matching models where all memories are stored in parallel, the continuous visual flow is inherently sequential. Our results show that the model can distinguish learned from unlearned route segments, with some tolerance to internal and external noise, including small displacements. The neural response can also explain observed behaviour taken to support sequential memory in ant experiments. However, obtaining comparable robustness to insect navigation might require the addition of biomimetic pre-processing of the input stream, and determination of the appropriate motor strategy to exploit the memory output.