Research

The big questions we get excited about

Around 600 million years ago, single cell animals started getting together in a big way. Over a relatively short period, an enormous proliferation of multicellular animals appear in the fossil record. The single celled animals split from plants around double that number of years ago. This fork in the road entailed two different strategies for the acquisition of energy: the "sit and absorb" strategy of the plants, and the "move around and get it" strategy of animals. Animal bodies are all about this strategy. Mechanical structures such as skeletons, actuation systems such as muscles, and sensing and control structures such as the nervous system, participate in making "move around and get it" the most successful strategy the planet has ever seen, leading to animals with baroquely complex, huge nervous systems that are able to visit other planets and knit colorful socks. We would like to understand the principles that make animals work. It's a problem that spans mechanics (musculoskeletal systems), control and sensing (nervous systems), and behavior most proximally, and ecology and evolution more distally. While many of most interesting challenges are in the realm of understanding the nervous system, we firmly believe that we can only understand the nervous system when we pay rigorous attention to the mechanical and behavioral sides of the issue, and we design our behavioral experiments with regard to what we know about an animal's ecology.

How we approach the big questions

We are committed to highly quantitative approaches that integrate engineering, simulation, and robotics with more traditional physiological approaches. The reason for this is that we believe analytical approaches to complex systems, where we break the problem down into simpler subproblems and model these, need to be complemented with synthetic approaches, where we build simulations and physical mechanisms, such as robots, to test our theories. A nice side effect of this commitment, given the fractal nature of biological systems where each subcomponent of a problem opens up into potentially lifelong research programs, is a type of “saliency filter” which says that if pursuit of some research problem isn't going to help fix a problem in our simulations or robots, then we probably won't be working on it.

From big questions to little problems

Currently, we focus on one animal for insights into these questions, an active sensing animal which senses its world through the emission of a weak electric field. This is the weakly electric fish, a fascinating animal that hunts in the dark through sensing perturbations of its self-generated field. This animal has been the subject of years of modeling efforts, allowing us to pursue more intricate issues. Why electric fish? As a fish, their brains are midrange in complexity (in terms of number of distinct neuronal cell groups in the brain, among other measures) between animals with very small nervous systems such as fruit flies or nematodes like C. elegans, which are also popular “model systems” in neurobiology, and animals with very large nervous systems, such as mammals. This affords certain advantages in terms of analysis. In addition, because they sense with electricity, it is quite convenient to generate and control input to the nervous system. Finally, mechanically they are relatively simple while being quite intriguing (for example they often swim backwards!). Thus, making connections between biomechanics and neural processing is facilitated. The active research projects on this system include

Computational estimation of sensing and motor volumes. We are applying the control-theoretic notion of the small-time reachable set to the analysis of how the sensing capabilities of the fish relate to its locomotor capabilities. Click this link to see 3D renderings of these volumes. We use optimal control approaches and idealized fish models for some of this work. Indications are that prey capture strikes follow energetically optimal trajectories through the fluid. Here is an animation that compares motion capture data of an electric fish (green) attacking a prey overlaid with an ellipsoid in an ideal fluid following the trajectory of minimum effort.
Neurophysiology of afferent signal processing. Experiments will soon be underway to examine how the brain processes afferent signals to extract features needed for guiding behavior. Our eventual target is midbrain areas where motor programs arise from the incoming sensory data.
Sensor based motion guidance. We are applying and extending modern probabilistic robotics theory to understand how weakly electric fish deploy sensors, movement, and control policies for hunting behavior, including search and capture. We are currently utilizing idealized mechanical models of the fish and realistic models of sensing as we work our way toward higher biological fidelity. Here is an animation of multisensor data fusion using an ellipsoidal approximation to the fish.
Reconstruction of neuronal input during natural behavior. Using a large simulation system we've created (AnimalLab), we reconstruct the input to all 14,000 sensory receptors on the body of fish during measured prey capture trajectories. We are utilizing these simulations for insights into upstream neuronal processing for extraction of musculoskeletal control signals.
Mechanisms of the novel propulsion system of weakly electric fish. Experimental methods include measurements of the propulsion capabilities of robotic models of electric fish. Theoretical methods include computational fluid dynamics to estimate the thrust of the ribbon fin and compare to the measurements from the robot. Here is a visualization of some early results using a modified immersed boundary algorithm.
Behavioral analysis of spatial cognition in active sensing animals. We've developed a hypothesis regarding why active sensing animals such as electric fish, bats, and rats tend to have highly accurate spatial maps of their surround. We are testing this hypothesis using fish mazes.

Some images and movies of our research can be found in the Uropatagium. Software and data to reproduce the results of more recent work (AnimalLab) is also available.

Larger image

Our research group pursues both empirical and modeling efforts in mechanics and neurobiology, integrating the two together within simulation environments.