Click here to login if you're an NAE Member
Recover Your Account Information
Author: Thomas Daniel
Interfacing Computer Electronics with Biology
Flight control in the hawkmoth is being analyzed by reverse engineering.
Animal movement emerges from the complex interplay of aerodynamic forces, nonlinear muscle forces, a massive flow of sensory information, and enigmatic information processing. In nearly all biological systems, the flow of sensory information is so massive that, at first glance, it appears to be redundant. We know, for example, that insects acquire sensory data for the strain distribution and position of a wing, chemical information from the environment around them, visual pattern data, and gyroscopic data. All of this information is integrated, stored, and processed by the insect’s nervous system. Interestingly, if one were to reach in and randomly remove a few neurons anywhere in this system, the chances are good that there would be no noticeable difference in the animal’s capacity to fly. By contrast, if one were to remove a few connections at random from the standard CPU design, the likelihood of functionality would be very low. This difference suggests redundancy in the design of neuronal systems, but our understanding of complex neural systems is still too primitive for us to draw that conclusion.
Overlying the issue of information processing is that the insect moves in spatially complex and uncertain environments. The flight of small insects in a naturally turbulent environment is particularly interesting in that the spatial and temporal scales of stochastic variations are large relative to the spatial and temporal scales of their control concerns. One of our central concerns is how biological systems deal with this combination of massive information flow and uncertainty.
To address this concern we have focused on a reverse engineering analysis of flight control in the hawkmoth, Manduca sexta. One of the fastest insect fliers, the hawkmoth is capable of navigating in low levels of light (even in starlight!), hovering while feeding on flowers moving in a breeze in turbulent eddies, and doing so with wings beating at about 25 Hz. How it controls its position in space, how it processes sensory information, and how we might affect direct computer-assisted control of its motion are central issues in our research.
We have, therefore, a challenging reverse engineering problem for a very successful, fast, tiny, self-replicating "robot" capable of flying autonomously, refueling in midair, and adapting to uncertain environments within milliseconds. From previous and parallel research, we know that these animals are inherently unstable devices that require a tremendous amount of very fast sensory information to control their position in space. Interestingly, the time demands occasionally exceed the rate at which data can be provided by the visual system. This suggests not only that fast information processing and transmission are extremely important, but also that parallel processing takes place between the slow, yet powerful, visual systems and the rapid, yet highly specific, mechanosensory systems.
The best way to picture the underlying control system is as a feedback loop between sensory information flow into the central nervous system and mechanical output in the form of a flight trajectory. Visual data pass to the central nervous system where motion detection is processed in visual centers of the brain. Their output is passed along a central nerve chord in the form of a motor pattern, some of which drives wing motions and some of which drives steering and compensatory motions in the abdomen and head. Moving wings propel the animal in ways that determine the flow of visual information; so too do steering motions from the abdomen; so too do head rotations (much the way our eyes track moving targets). By this scheme, the loop between visual data flow and motion is closed.
Insects spend a great deal of time processing images - cell-by-cell - and a great deal of innovative research is being done, largely by postdoctoral researchers and graduate students, to understand this processing in detail. For example, some cells in the visual processing centers of the brain of hawkmoths have been shown to be tuned to a specific direction and velocity of movement in the visual world of the insect and that this direction and velocity tuning appears in the motor pattern in the central nerve chord. In addition, muscles responding to these motor outputs operate in a nonlinear manner in ways that mimic mammalian cardiac cells. Like hearts, insect flight muscles commonly drive a structure in which elastic energy storage plays a key role in how motion is maintained and controlled. These muscles actively deform the thorax, which, indirectly, drives the up and down motions of the wings. In moths, the thorax muscles are activated for each wing stroke, sometimes reaching frequencies in excess of 30 Hz. In other insects, muscle activation is less coupled, with many wing strokes following each stimulus. In mosquitoes and fruit flies, for example, wing beat frequencies can exceed 200 times per second - much higher than the rates at which muscles can turn on and off - and follows from a resonant oscillation of the thorax to which wings are attached.
To explore this complex control system, we use a wide variety of experimental and theoretical approaches, from intracellular recordings to computational models of aeroelastic behavior of oscillating wings and analyses of quasisteady aerodynamic flight forces. High-speed digital videography, combined with simultaneous recordings of neural information during flight, links these experimental and theoretical approaches and sets the stage for a formal systems analysis of the control circuit of flight.
In our project, we hope to equip freely flying insects with onboard microcomputers (i.e., programmable systems-on-a-chip) that can track neuronal systems and enable us to eavesdrop on the control circuit. In theory, microcomputers can also send digital information to neural centers of the animal where local processing may enable us to delve more deeply into the feedback control of these very successful "robots."
Although this study heralds an exciting new approach to the reverse engineering of biological systems, we remain mindful of the challenges that lie ahead. Can we use the digital world of CMOS electronics to control motion in biological systems? Do neural systems adapt to such digital control events? Can we implement learning algorithms in implantable microcomputers? New data suggests that we can make significant progress toward answering these questions.