BIOINSPIRED SOFT ACTUATORS FOR EYEBALL MOTIONS IN HUMANOID ROBOTS: DESIGN, DEVELOPMENT, AND CONTROL

Abstract

Technological progress in advanced materials and artificial intelligence has given an impetus to the development of humanoid robots that can mimic human appearance and interact with people effectively. Eyeball motions play a key role in facial expression of humanoid robots, and can greatly improve the relationship between the robots and the users. This thesis makes several contributions in the design, development, and control of soft robotic eyes. Two designs for soft robotic eyes are developed, the first based on dielectric elastomer actuators and the second based on pneumatic actuators. Two model-free control methods by exploiting reinforcement learning are proposed, the first based on Q-learning and the second based on deterministic policy gradient. This thesis develops robotic eyes that can mimic natural human eye movements and track moving objects accurately. It is expected that the contributions from this thesis would be useful in humanoid robots with enhanced human-robot interactions.