UC Berkeley

Dynamic locomotion behaviors in both animals and machines emerge from high-dimensional, nonlinear, and coupled interactions between neuromuscular or actuator feedback control and the mechanical interactions between the body and the surrounding environment. The complexity of the hierarchical control systems and mechanical behaviors of multiple distributed appendages and body segments necessary to achieve robust locomotion behavior in challenging environments has made elucidation of general principles underlying the remarkable adaptive and dynamic locomotion behaviors observed in animals, and consequently the achievement of such behaviors in engineered systems, a significant challenge. In this dissertation, we take the first steps towards developing a principled theory and practice for understanding mechanisms of whole body dynamic locomotion control through the hierarchical composition of distributed embodied control modules. Specifically, we introduce a data-driven approach for discovery of hierarchical neuromechanical control systems via the targeting of a low dimensional posture principle, demonstrate how the embodiment of multifunctional actuation affordances and recruitment of multiple body segments enables robust, high performance dynamic ground righting maneuvers, and show that the sequential composition of substrate manipulation modes can enable deep, rapid burrowing in small organisms.

In Chapter 1, we present an approach for discovering task-specific attracting invariant postural submanifolds directly from ensemble behavioral measurement data. Specifically, our approach is grounded in the templates and anchors hypothesis for dynamic locomotion control, which hypothesizes that the preferred postural degrees of freedom used for a specific task are anchored - rendered stable and robust to perturbations - by fast neuromechanical feedback control forces. We present both a theory and practice for identifying these dynamic posture principles that depends only on a local model of the hypothesized template-anchor dynamics, and therefore is generalizable to a wide range of locomotion phenomena including both periodic and transient behavior. We introduce an expressive model of periodic locomotion behavior, the generalized anchored Hopf model (GAHM), that affords direct numerical tests of the efficacy of our approach. We find that our data-driven method is capable of accurately detecting template submanifolds with a surprisingly wide range of geometric and dynamical properties, and is robust to measurement noise and parameter errors.

In Chapter 2, we explore how the composition of multifunctional actuation affordances can enable robust dynamic transitional maneuvers. Specifically, we use the ground righting behavior of the house gecko H. platyurus to explore how the composition of mechanical affordances embodied in the tail and spine may enable robust performance in response to different environmental contact conditions and tail autotomy. We develop a parsimonious axial plane ground righting template model capable of predicting righting times and trajectories for the rolling behavior of bodies with different effective axial plane shapes, allowing us to test the hypothesis that control modules beyond tail driven actuation alone contribute to ground righting performance. To reconstruct the axial plane ground righting kinematics of multiple body segments, we measure the full 3D spatial kinematics of landmarks on the tail, torso, and legs of geckos ground righting on substrate conditions either allowing or preventing tail-ground contact. We discover comparable total ground righting times across distinct tail actuation modes, with performance benefits afforded when the tail is employed in contact with the substrate. Further, we find that in individuals with the tail autotomized, ground righting can still be achieved using dynamic spine motions with comparable performance to tail-intact individuals. Finally, we take the first step towards exploring dynamic-spine tail compositions by showing that measured ground righting performance exceeds the predictions of the passive template model, and that a shared axial plane spine twisting control mode is present for all tail actuation strategies.

Finally, in Chapter 3, we explore how adaptive appendage control can enable rapid, deep burrowing in complex multiphase substrates for small, force-limited organisms. Specifically, we use the Pacific mole crab, E. analoga, to study how leg kinematics adjust in response to depth-dependent substrate resistive forces encountered during burrowing. To test the hypothesis that appendage control is adjusted in response to increasing substrate resistive forces, we use optically transparent substrates with significantly different resistive properties to precisely measure the effects of body depth and substrate resistance on appendage kinematic control. We find that the stride periods of the anterior excavating appendages increase as substrate resistive forces increase. We further find evidence suggesting the presence of an additional substrate manipulation mode, legged cavity expansion, that may be sequentially composed with excavation strides to enable burrowing to increased depths. Finally, we show that the control program of the anterior excavating appendage pairs is robust to loss of function of a posterior appendage pair, and take the first step towards identifying the adaptive, sequentially composable substrate manipulation modules that enable burrowing using particle image velocimetry (PIV) measurements of the substrate mechanical response.