Date of Award

5-2017

Document Type

Thesis

Degree Name

Master of Science (MS)

Legacy Department

Electrical and Computer Engineering

Committee Member

Dr. Ian Walker, Committee Chair

Committee Member

Dr. Robert Schalkoff

Committee Member

Dr. Adam Hoover

Committee Member

Dr. Paul Yanik

Abstract

Traditional robots are constructed from rigid links which facilitate both stiffness and accuracy. However, these systems operate best in open, highly structured spaces, and environments traversable by this technology are inherently restricted to scales and geometries which match the size and shape of the links. Conversely, continuous backbone continuum robots have enormous potential for adaptive exploration of unstructured environments. However, to date there has been very little research on algorithms for learning and adapting to changes in environmental conditions with continuum robots. In this research, we introduce new results in learning policies for novel long, thin, continuously bending continuum tendril robots aimed toward applications such as remote inspection and sensor mobility for improved sample acquisition. The results could also have potential applica tions in defense and security, search and rescue in hazardous environmental conditions, and as an innovative option for sensor placement in environmental monitoring. Using a prototype continuum tendril robot previously developed at Clemson University, we demonstrate the new learning policy for the tendrils adaptive sensor placement and remote inspection within an environment seeded with numerous disparate and slowly (over a matter of hours) time-varying sources, and discuss the potential for use of such robot tendrils in environmental monitoring applications. The learning algorithm implemented in real-time is shown to help the tendril to adapt its sensor placement to changing environmental sources.

Share

COinS