A Novel Continuum Manipulator
Abstract
We address the problem of controlling continuum manipulators and evaluate Reinforcement Learning to produce a control policy for a robotic platform. Our approach discretizes the state and action spaces to reduce the training needed to converge to an optimal policy. We integrate Q-Learning, computer vision, and a pneumatic system into a single robotic platform. The agent is tasked with tracking and striking a target with a continuum manipulator modeled by a party-blower. We describe Reinforcement Learning, the methods used to train the agent, and describe the performance of the optimal policy successfully striking the target.
Subject
Reinforcement LearningContinuum Manipulator
Artificial Intelligence
Robotics
Machine Learning
Computer Science
Citation
Cochrane, Eric C (2017). A Novel Continuum Manipulator. Undergraduate Research Scholars Program. Available electronically from https : / /hdl .handle .net /1969 .1 /164392.