Data driven modeling of behavior

With a quickly growing literature on insect behavior it is challenging to understand how individual behavioral “modules”—e.g. zigzagging upwind to follow odor plumes, following visual and thermal cues, and using visual information to decelerate and land safely (see Figure)—interact in natural environments. To organize this knowledge, we are working on bringing together control-theoretic and machine-learning tools to build data-driven open-source models of insect behavior. The methods we are developing will be equally applicable to extracting feedback control models for any type of biological, or robotic system.

GraphicalAbstract.jpg

Image Caption: To locate human hosts, mosquitoes use seven behavioral “modules”, described above. Each of these behaviors is independent in that they can occur in different orders, depending on the order in which they encounter odors, visual cues, and thermal signatures. The behaviors are, however, indirectly linked to one another through the animal’s interaction with its environment. For example, approaching visual features increases a mosquito’s chance of encountering a thermal plume. These indirect linkages lead to nonlinear interactions, making it difficult to predict what insects will do in novel environments, or how the neural circuits underlying their behavior might interact in the natural world. Graphic from our paper on how mosquitoes integrate sensory modalities.

Funding: Moore & Sloan Foundations, Washington Research Foundation, Sackler Scholarship


Long distance migration of Drosophila melanogaster

Thirty years ago ecologist Jerry Coyne and collaborators demonstrated that flies are capable of flying over 10km; we are working on replicating these results with technologically advanced traps to find out how. Work in progress in collaboration with Kate Leitch and Michael Dickinson at Caltech.

Funding: Simons Foundation, NSF

 Our collaborators: Kate Leitch, Francesca Ponce, and Michael Dickinson, out on the playa, examining a fly trap.

Our collaborators: Kate Leitch, Francesca Ponce, and Michael Dickinson, out on the playa, examining a fly trap.

 The great release - thousands of hungry flies to take the air.

The great release - thousands of hungry flies to take the air.

 We chose to work on this desert playa because of the lack of visual and olfactory cues.

We chose to work on this desert playa because of the lack of visual and olfactory cues.


 An Alkali Fly (Ephydra hians), under water at Mono Lake, safely wrapped in a protective air bubble.

An Alkali Fly (Ephydra hians), under water at Mono Lake, safely wrapped in a protective air bubble.

An Alkali fly (ephydra hians) walking under water at Mono Lake, CA.

Alkali Flies of Mono Lake

In late summer, the shores of Mono Lake, California, are bustling with small flies, Ephydra hians, which dive under water inside small air bubbles to feed. The best description of their behavior is offered by Mark Twain, in his book Roughing It:

“You can hold them under water as long as you please–they do not mind it–they are only proud of it. When you let them go, they pop up to the surface as dry as a patent office report, and walk off as unconcernedly as if they had been educated especially with a view to affording instructive entertainment to man in that particular way.”  – Mark Twain, 1872

Using a combination of high speed videography, force measurements, scanning electron microscopy, and manipulations of water chemistry we found that these flies are uniquely adapted to staying dry in the highly alkali waters of Mono Lake thanks to their extra hairy bodies.

Publication:  van Breugel, F and Dickinson, M. Superhydrophobic diving flies (Ephydra hians) and the hypersaline waters of Mono Lake. (2017) PNAS.

Press: BioGraphicWashington PostNational GeographicNew York TimesNew ScientistSciencePopular ScienceReutersGizmodoMercury NewsDaily MailNewsweekIFL Science, Vice

Funding: National Geographic Committee for Research and Exploration


How flies use CO2 to find food

The olfactory system in Drosophila is a key model system for studying both innate and learned behavior in animals. Within this framework, CO2 has become the canonical odorant for studying innate aversive behaviors. However, this result goes the natural intuition that a fruit fly, which is attracted to CO2-emitting fermentation processes, should in fact be attracted to CO2.

We used automated computer vision systems to collect over 50,000 fly-hours of data to finally resolve this longstanding paradox. We found that flies only exhibit aversion to CO2 in relatively unnatural conditions like those imposed by the popular T-maze assay. When flies are allowed to acclimate to their environment, and are in an active foraging state, they exhibit strong attraction to CO2. To provide confidence in our results, we repeated our experiments in four independent behavioral assays including traps, free flight studies, a landing platform, and constrained walking arenas.

Publication: Distinct activity-gated pathways mediate attraction and aversion to CO2 in Drosophila.
van Breugel, F., Huda, A., and Dickinson, M. (2018) Nature.

Funding: NIH and the Simons Foundation.


 Long exposure of an illuminated fly as it approaches a fermenting strawberry.

Long exposure of an illuminated fly as it approaches a fermenting strawberry.

How insects follow odor plumes

When flies, mosquitoes, and other insects encounter an attractive odor, they turn upwind. Because odor plumes are broken apart by turbulent flows, the insect invariably exits the plume, sometimes after just a few milliseconds. This triggers zigzagging back and forth, until they re-encounter the odor plume. This strategy generally leads them close to the odor source, however, visual and other cues are necessary for the final stage of search. 

Publications:

van Breugel, F., Riffell, J., Fairhall, A., and Dickinson, M. H. Mosquitoes use vision to associate odor plumes with thermal targets. (2015). Current Biology.

van Breugel, F. and Dickinson, M. H. Plume-Tracking behavior of flying Drosophilaemerges from a set of distinct sensory-motor reflexes. (2014). Current Biology.

PressBBC (mosquitoes)

Funding: NIH, NSF, AFOSR, Hertz Foundation, Paul G. Allen Family Foundation


Visual control of flight & landing

Using 3D tracking, closed loop control of high speed cameras, and genetic tools, we explored landing and the neural basis for flight speed control of Drosophila in free flight. Inspired by my results, we developed a novel algorithm for distance estimation from a single camera using nonlinear control theory.

Publications:

van Breugel, F., Suver, M. P., and Dickinson, M. H. Octopaminergic modulation of the visual flight speed regulator of Drosophila(2014). J. Exp. Biol.

van Breugel, F., Morgansen, K. A., and Dickinson, M. H. Monocular distance estimation from optic flow during active landing maneuvers. (2014). Bioinspiration and Biomimetics.

van Breugel, F. and Dickinson, M. H. The visual control of landing and obstacle avoidance in the fruit fly, Drosophila melanogaster(2012). J. Exp. Biol.

PressBBC (landing)

Funding: NSF, AFOSR, Hertz Foundation

 Photo montage of a fruit fly landing.

Photo montage of a fruit fly landing.

A fruit fly comes in for a landing, filmed at 5,000 frames per second.

 Strobe-light illuminated image showing both the inward and outward wing strokes of this 8-winged flapping hovering MAV (2008).

Strobe-light illuminated image showing both the inward and outward wing strokes of this 8-winged flapping hovering MAV (2008).

Bio-inspired design of flYing robots

As an undergraduate, I used genetic algorithms and simulations to design wing stroke patterns for flapping flight. Subsequently, I built physical flapping systems, culminating in the first passively stable flapping hovering machine (left).

Publications:

van Breugel, F., Regan, W., and Lipson, H. From insects to machines. (2008). Robotics and Automation Magazine.

Regan, W., van Breugel, F., and Lipson, H. Towards evolvable hovering flight on a physical ornithopter. (2006). Alife X conference proceedings.

van Breugel, F., and Lipson, H. Evolving buildable flapping ornithopters. (2005). GECCO conference proceedings.

Funding: Cornell Presidential Research Scholars