[back]

The Neuroethology of Sensory-Based Behavior
Malcolm MacIver, Joel Burdick

Fish Sensing
We have begun research that addresses the interrelationship between animal sensing and the mechanics of animal movement. There are two interrelated thrusts to this work. The first is optimal sensing and movement strategies for far-field targets, such as distant resources that must be detected and acquired. This work further diverges into two tracks, the first concerning how arrays of weak signals are efficiently filtered in order to extract the necessary control parameters for the subsequent behavior; the second concerning whether movement strategies that we have previously quantified for one particular animal are time-optimal, energy-optimal, sensory information-optimal, or some combination of these.

The second thrust is optimal sensing and movement strategies for near-field locomotion-directed signals, such as needed for sensing flow velocity near a constriction in a stream-bed that requires a fish to increase its thrust. Here, we seek to understand some of the bases of the extraordinary maneuverability and efficiency of animal movement, with evidence from fish and insect locomotion indicating that near-field sensing of the surrounding flow may be integral to these very desirable properties.

Figure 1. Large scale simulation of the 15,000 sensory receptors located on weakly electric fish during prey capture behavior. These simulations allow us to quantify the sensory information that corresponds to movement patterns.

While our own experience might suggest a model of "sense then act," animals appear to operate at such a low level of signal strength that their movement is typically part of their sensing performance and strategy. This insight has been pursued within the engineering domain under the rubric of "active sensing," most vigorously within the field of active vision. The difference between this style of research and prior theoretical approaches to perception is the emphasis on movement as fundamental to the act of sensing. For example, David Marr begins his well-known 1982 book on vision with the statement that "vision is the process of discovering from images what is present in the world, and where it is"; this is what active vision researcher Andrew Blake called "a prescription for the seeing couch potato" (1995). In contrast, in the active sensing view, behavior is tightly coupled to sensing, and behavioral programs operate on minimalist representations of the world that are computed from changes in the sensory information reaching the animal as it manipulates its body, and thus its biological sensor arrays, through space. Thus, behavior is no less dependent on sensing than sensing is on behavior.

A common theme to both thrusts of our work, target-directed far-field sensing and movement and locomotion-directed near-field sensing and movement, is to utilize more abstract approaches to animal movement from geometric mechanics to understand what information is needed to support movement. In this "info-mechanical" approach, we seek a unique synthesis of modern neurobiology of sensory systems with ongoing research in geometric mechanics that unites control theory with mechanics for understanding the behavior of underactuated mechanical systems (where the possible ways the object can move is greater than the number of thrusters that can generate this movement) such as animals. For example, work in geometric mechanics and control theory suggests that for any underactuated system to move to a location in space, periodicity in the control variables is necessary. This places work on the neural central pattern generators that are responsible for terrestrial, aerial, and aquatic movements in animals on a new theoretical level, and should lead to some fundamental insights into the complementarity of animal body plans and their neuronal control systems.