Research

How does a brain decide which memory to use? I work on that in Drosophila, combining closed-loop behaviour, two-photon imaging and circuit-level genetics — and, where it helps, models simple enough to run in a browser.

Now

Postdoctoral work in the Felsenberg Lab: how visual context decides which olfactory memories are retrieved.

Paradigm diagram: a contingency table pairing two odours with a green danger panorama and a blue safe panorama, a fly walking on a ball inside a closed-loop visual arena, and a 2D grid of alternating odour patches. — **Neuronal mechanisms of context-dependent memory retrieval** A fly learns that an odour predicts a targeted infrared laser pulse — but only in one visual context. I built a closed-loop virtual reality to put that context under experimental control, designed so the mushroom body can be imaged while the fly walks through it.

Before

PhD work on the circuits and behaviour of foraging flies, and earlier work on insect-inspired navigation in walking robots.

Graphical abstract: a fly that has visited food, with metabolic state driving locomotor adjustments and sensory guidance feeding into short and long trips, beside a trajectory whose loops widen over time into an expanding search pattern. — **What shapes a fly's search after it finds food** A fly that finds food does not walk away from it — it searches the area it just left. How long it keeps searching, and how far it ranges, depends on what it has recently eaten and how sated it is, so the pattern widens as the memory of the meal fades. Curr. Biol. 2026 →

Confocal image of a fly brain with one neuron of a bilateral pair labelled in magenta and the other in cyan, each running from a cell body at the outer edge to dense arborisations, meeting along a shared tract at the midline. — **Turning hunger and patch richness into a decision** How long a fly stays at a patch of food depends both on how hungry it is and on how rich the patch is. This work traces the neurons that carry those two estimates and turn them into a decision to stay or move on. bioRxiv 2023 →

Duotone illustration: a hand shines a torch across a table of foods — oatmeal, chocolate, tofu, greens and a doughnut — picking some out of the dark and leaving others in shadow. — **optoPAD** A closed-loop system that detects the moment a fly starts to feed and fires a light pulse within milliseconds, so neurons can be activated while the animal eats rather than at some arbitrary time. It turns optogenetics from something done to a fly into something done during a behaviour. Illustration: Diogo Matias, Champalimaud Centre for the Unknown eLife 2019 →

Model schematic: head-direction, home-vector and global-vector neuron arrays feeding an action-selection stage and a CPG-based locomotion controller, beside the path-integration geometry between nest and feeder. — **Adaptive vector navigation in embodied agents** A neural model of path integration that lets a walking agent forage away from its nest and return along a home vector, without a map. Front. Neurorobot. 2017 →

Learning-circuit diagram beside a simulated hexapod robot climbing a step, and photographs of the physical AMOS hexapod performing the same climb. — **Adaptive obstacle negotiation in embodied agents** A hexapod robot that learns to anticipate obstacles from its own sensory history rather than being told the terrain in advance. Front. Neurorobot. 2014 →

Two photographs of a person wearing a powered lower-limb exoskeleton, once climbing stairs and once seated. — **Lower-limb exoskeleton** Early engineering work on a powered lower-limb exoskeleton, before I moved to neuroscience.

The full list of papers is on the publications page.