![](/wp-content/uploads/2021/03/abstractheadergraphic2013.gif)
Meeting Abstract
114.5 Monday, Jan. 7 Using reduced-order models to study dynamic legged locomotion: Parameter identification and model validation BURDEN, S.A.*; REVZEN, S.; MOORE, T.Y.; SASTRY, S.S.; FULL, R.J.; Univ. of California, Berkeley; Univ. of Michigan; Harvard Univ.; Univ. of California, Berkeley; Univ. of California, Berkeley sam.burden@berkeley.edu
Generating testable hypotheses for dynamic legged locomotion is challenging because motion imposes a continually-changing reference frame, and perturbations typically induce nonlinear effects. Fortunately, rhythmic biological motion is often highly stereotyped and low-dimensional, suggesting amenability to description by reduced-order dynamical models as proposed by the Templates and Anchors Hypothesis (TAH). Such models can predict experimental outcomes that cannot otherwise be quantified. For instance, during perturbations from the environment, purely mechanical self-stabilizing behavior can be defined, so that deviations resulting from neural feedback can be explored. However, given a candidate reduced-order model, there is seldom a direct method to measure free parameters and validate the model. Operationalizing TAH requires statistical tools to estimate parameters and select models using data collected within experimental paradigms. We propose a computationally-tractable method for applying nonlinear regression to the piecewise-defined dynamical models that naturally describe terrestrial locomotion. We illustrate the technique using data from an experiment involving center of mass (COM) perturbations and mass distribution manipulations applied to running cockroaches. Preliminary results corroborated our initial finding that neural feedback could be delayed by 1-2 strides after perturbation onset and demonstrated that a parsimonious spring-mass model for horizontal plane dynamics of sprawled running animals (Lateral Leg Spring) provides an accurate quantitative prediction of the animal’s COM dynamics during this interval. Our approach can be applied more generally to dynamical systems ranging from muscles to swarm coordination.