How do nervous systems transform sensory signals into motor commands to guide locomotion? To address this question, this research examines a class of sensory guided locomotion tasks uniquely amenable to computational and engineering analyses: sensorimotor stabilization tasks. In these tasks, animals automatically modulate muscle commands to drive sensory signals (e.g. electrosensory, tactile, visual) to desired equilibria or limit cycles.
Specifically, this research examines three remarkably divergent animal locomotor behaviors: multisensory control of swimming in weakly electric knifefish, high-speed antenna-based wall following in cockroaches, and visual yaw control in fruit flies. While performing sensory guided locomotion tasks, these animals are subjected to behaviorally naturalistic perturbations that lead to rich transient dynamics during recovery. Mechanical signals (e.g. positions, forces) and neural activity (e.g. action potentials) during recovery to these perturbations are used to validate (or refute) specific closed-loop sensorimotor control models. These models identify the roles of mechanics versus neural computation in the stability and performance of each behavior. The same approach to modeling, analysis, and experimentation is vetted in a biomorphic robot, establishing an experimental baseline in a highly controlled context, as well as enabling the translation of biological control strategies to a robotic platform.
This research, disseminated broadly through both the engineering and biological literatures, lays a scientific foundation on which to develop biomorphic robots for critical applications such as disaster recovery, space exploration, and security. Longer term, the unified approach to biological and robotic modeling developed in this project may lead to enhanced neural prostheses and brain--machine interfaces.
This Presidential Early Career Award in Science and Engineering (PECASE) was instrumental in creating a transformative approach to the analysis of animal and robot movement and its control. Here "control" refers to mechanisms like a thermostat or cruise control, where a variable is measured (e.g. temperature or speed), interpreted (e.g. "too cold/hot" or "too fast/slow"), and used to regulate a system input (e.g. furnace or gas pedal). There is a surprisingly large gap between running, swimming, and flying animals and their engineered robotic counterparts. One key reason for this is that we don't yet understand how to control robotic movement. This research addresses basic underlying scientific questions about control of movement in animals and machines in an attempt to close this gap. The PI set out to create tools and approaches that would enable researchers to model an animal's control strategies, making it possible to translate them into algorithms for robots. The work has resulted in dozens of papers and presentations at international conferences and has led to new algorithms for improving the stability and agility of robotic systems. Several news articles have been written about various aspects of this work that are accessible to non-specialists: http://eng.jhu.edu/wse/magazine-summer-12/item/bio-bots/ http://jeb.biologists.org/content/216/9/i.2.full.html?etoc www.bbc.com/news/science-environment-22130854 www.futurity.org/double-play-motion-keeps-critters-stable-agile/
