Skip to content Skip to navigation
Masters Defense
9/28/2017 02:00 pm

Designing and Learning CPG Gaits for Spherical Tensegrity Robots Using Bayesian Optimization

Colin Rennie, Dept. of Computer Science

Defense Committee: Prof. Kostas Bekris (Chair), Prof. Abdeslam Boularias, Prof. Konstantinos Michmizos


This work presents a framework for developing a library of gaits for tensegrity robots, which are examples of highly non-linear, hyper-redundant, compliant systems. The first component corresponds to the design of a Central Pattern Generator (CPG) for such robots, which provides a reparametrization of the system that easily results in the generation of rhythmic gaits. Second, a novel framework is presented for simultaneously discovering effective gait parameters along different directions of motion. The framework integrates a parallel Bayesian Optimization (BO) process with classification. This integration is more efficient than Monte Carlo sampling or BO or classification alone. Evaluation is performed in simulation using a spherical tensegrity robot