Robotics has drawn inspiration from nature for many years, but only recently has an understanding of the musculoskeletal dynamics of animal running been successfully implemented in small, self-stabilizing legged robots. One such example is the biomimetic hexapod Sprawlita, capable of running at over 2 bodylengths per second and traversing hip-height obstacles, all without feedback control.; Motivated by the question of how these robots can take advantage of feedback information, this thesis explores sensory-based cyclic dynamic tasks---in animals, legged robots, and dynamic systems in general---toward understanding the mechanisms and functional roles of sensory feedback and a general approach for designing adaptive controllers.; To explore sensory-mediated cyclic behaviors in animals, cockroaches running on an inertial treadmill are subjected to sustained oscillatory perturbations and electromyograms are recorded to determine a phase measure relative to the perturbation. The cockroach motor pattern generators are modeled using a feedback coupled nonlinear oscillator. The observed behavior is consistent with this model and a statistical test is used to quantify the phase measure distributions.; Feedback coupled nonlinear oscillators as controllers for cyclic dynamic systems are then examined, focusing on designing for adaptation to changing environmental conditions. An existing visual design method is expanded upon, creating an intuitive three dimensional representation of the plant and the non-linear oscillator controller as conditions change, visually predicting the coupled system adaptive behavior. An analysis tool, the omega contour analysis, is developed and used to determine the appropriate feedback characteristics. Additionally, a new method for specifying the coupled system behavior using intentional time delays is presented.; Finally, this thesis concludes by developing an adaptive controller for a numerical simulation of the robot Sprawlita. With the biologically-inspired adaptive controller, the robot runs up slopes 33% faster than in the open-loop configuration. Further, additional design tools are developed for feedback coupled nonlinear oscillator systems with binary actuation and feedback.; This thesis outlines a biologically-inspired approach for achieving adaptive behaviors. The design and analysis tools developed are general, and can be used to design a feedback coupled nonlinear oscillator controller for any cyclic dynamic system.
展开▼
