Janelia Research Campus
Dr. Reiser is a group leader at the Janelia Research Campus.
Drosophila Flight Control
Michael Reiser vividly remembers the clunky mechanical muse that sparked his fascination with neuroscience as an undergraduate. The gadget was a color-seeking mobile robot he cobbled together from a broken laptop, a webcam, some proximity sensors, and other parts scrounged from a University of Florida engineering lab.
"You could command the robot to seek 'red,' and then you could hold a Coke can in front of it, and it would follow you around the room," he says. In turn, that robot helped lure Reiser into studying "multisensory fusion"—how an organism, such as a fruit fly, integrates information from sight, smell, hearing, and several other senses to produce a suitable response to its ever-changing world.
"It's a fundamental problem that is not particularly well studied or understood," he says. A theoretical understanding of biological sensor fusion remains elusive, but clearly the nervous systems of critters from insects to people have solved this huge challenge, says Reiser. "You would think that the more sensors you have, the better your robot is going to function, because it knows more about the external world. But, it turns out that as you, say, double the number of sensory signals coming into your system, at the very least you'll need to quadruple the lines of code that analyze and respond to all this sensory information."
Such a massive increase in input presents difficult control problems, says Reiser, because "all kinds of weird combinations can happen. Sensory information might interact in strange ways, with one sensor telling you you're about to crash and the other one saying not."
As a master's degree student at the University of California, Berkeley, Reiser was introduced to the challenge of understanding fruit fly flight as a model for the neural control of behavior. In Michael Dickinson's laboratory, a multi-disciplinary group focused on the neurobiology, aerodynamics, and behavior of flies, he constructed a fly-inspired robotic control system as part of a project to ultimately build a fly-sized flying robot. Importantly, says Reiser, he also schooled himself in machine learning and control theory, which enabled him to build rigorous computational models of the biological phenomena he was exploring. When the Dickinson lab relocated to the California Institute of Technology, Reiser continued his research in the control of fly flight, while seeking his doctorate in the Computation and Neural Systems program.
"The fly represents a 'Goldilocks' animal, in that it is 'just right,'" he says. "It's not so ridiculously simple that you can't observe complex behaviors. But it's also not so complex that you don't actually have some tractable approaches to working on it."
At Caltech, Reiser used his engineering background to design an LED-based modular display system that is used as a virtual-reality flight simulator for tethered flies. "Basically, it's like a set of programmable Legos that you can snap together in different configurations to make different visual environments for the flies," he says.
While other researchers had long studied how tethered flies responded when the visual world around them is rotated, Reiser literally introduced a new dimension to the investigation. He set out to explore the flight response of flies to "translational" visual stimuli.
"There's a lot of interesting and important work on the response of flies just to turning," he says. "But the geometry of vision when you add translational movement is far more complicated. When a fly turns, the only visual information it can use is an estimate of how fast the world is rotating around it. This estimation does not depend on the structure of the world around the animal. But as soon as the fly flies forward, the visual world becomes much more complicated. Structure emerges that the animal must respond to so that it can avoid collisions and navigate in a sensible way."
One approach that Reiser thinks will be critical for advancing studies of fly sensory control is the "closed-loop" experiment, in which the fly's flight response to its environment is fed back to trigger further changes in the environment. Such closed-loop studies, says Reiser, are often a more appropriate stimulus for the animals' flight control systems than traditional open-loop methods, which expose a fly to a controlled stimulus and measure its response.
"We are constantly trying to come up with new types of closed-loop experiments since they tend to be very informative," says Reiser. "The animal's nervous system did not evolve for open-loop conditions. When a fly attempts to turn to the right, there is an expectation that the visual world will move to the left. When an experiment breaks this loop, we should think carefully about how this condition differs from what a freely flying fly would experience."
Using a combination of open- and closed-loop methods, Reiser has worked out an algorithmic description of how a fly's nervous system processes visual information in the face of these translatory visual conditions. He has recently built a programmable gantry that can whisk a tethered fly through a precise trajectory, to better stimulate the sensory systems that respond to motion—continuing to increase the fidelity of these virtual-reality simulations.
At Janelia, Reiser plans to use his programmable, closed-loop, multisensory environments to analyze in detail how flies use multisensory inputs to control flight. He also hopes to work with Janelia experts in fly genetics and molecular techniques to zoom in on the detailed neuronal circuitry underlying flight control.
"To make progress on understanding the nervous system, you certainly need to study it from the 'bottom up'—using the techniques of molecular biology, electrophysiology, and sophisticated microscopy," he says. "But we should couple these approaches with a 'top-down' analysis of how the information that is coursing through that animal's nervous system plays out as its behavior. It's not clear to me that either approach alone will ever be wholly satisfying, because ultimately we want a full understanding of how the nervous system implements algorithms that make complex behaviors possible," he says.