Skyline Navigation

My Phd involves building navigation systems inspired by the visual system of desert ants that forage for food during the hottest part of the day. Due to the high environmental temperatures they are unable to use pheromone trails and resort to visual navigation and path integration (dead reckoning) to find their way back to the nest.

Desert ant eyes have a wide field of view and are sensitive to UV and green light; attributes that allow them to extract the skyline very effectively despite lighting conditions and cloud cover. This leads to the hypothesis that the ants could be using the shape of the skyline to navigate.

Edinburgh castle in UV.
Edinburgh Castle photographed in UV (Stone 2014).
Notice how easily separable land and sky are.
My talk at RSS 2014 on skyline for navigation.

A proof of concept navigation system was demonstrated by capturing panoramic images in UV along a route in Edinburgh. Using SeqSLAM these images could be used to localise successfully despite being very small binary images. Follow-up work showed that some extra benefits fall out of representing the scene in this way, including 3-axis tilt invariance if frequency representations are used.

View details »

Interactive poster at ICRA 2016 on tilt invariant localisation.

Central Complex Modelling

The central complex is a part of the insect mid-brain that is remarkably conserved across species and highly implicated in path integration. Part of it (the protocerebral bridge) includes a solar compass. Using connections and spiking activity observed through electrophysiology experiments, I am working on a proof of concept neuronal network that implements path integration.

Poster describing a potential path integration circuit in the insect central complex.
(won Best Poster prize at Janelia Farm Central Complex IV conference)

Lemma

Lemma is a website that can be used for exploring arguments that was created for a 3 Day Startup event. I pitched the initial idea and then led the group project where we developed it further, using Rails and Ember to create a cool proof of concept site.

All arguments are stored in a graph data structure, making use of argumentation theoretic concepts (premises, conclusions) to link them. The graph can be explored, so that any aspect of an argument can be scrutinised. This allows you to explore by clicking through our graph, just like wikipedia but for opinions. It also allows you to pinpoint weak points of any arguments.

Backdrop video to 3 Day Startup presentation of Lemma (silent).