Building and Controlling Fluidically Actuated Soft Robots: From Open Loop to Model-based Control
Compared to traditional robots, soft and compliant robotic designs are well-suited for manipulating objects and interacting with the environment because they reduce the need for accuracy and precision in motion. My thesis work contributes to the field of soft robotics by tackling the question of how to intelligently integrate compliance into robot design as well as how to fabricate and control soft robots for many tasks. I demonstrate how model-based control, enabled by my advances in soft fabrication and modeling, can provide for dynamically actuated soft robotic manipulation and locomotion systems.
My thesis advances soft fabrication through the development of casting and printing techniques, novel actuator morphologies, and embedded actuation modalities. Models developed for soft robots enable object identification through deformation and dexterous manipulation under contact. Control algorithms developed for soft robots realize open-loop locomotion control, closed-loop manipulation, grasp planning, and model-based control.
These techniques are combined in designs that include a soft robotic fish, a soft robotic hand, a soft juggling robot, and soft robotic arms. I use modular soft designs to create robotic fish for underwater locomotion and proprioceptive hand designs and manipulator arms for object manipulation. I develop a cyclic pump systems for undulating soft actuators and an embedded system design process to create an autonomous soft fish robot that receives remote acoustic commands to explore oceanic life. A proprioceptive soft robotic hand grasps a wide range of objects and performs haptic object identification by conforming to their shapes. Dynamic object manipulation with a soft surface is enabled through modeling soft contacts and by applying model-based control. Closed-loop controlled manipulator arms perform autonomous grasps, move in confined spaces and perform manipulation under contact. Model-based control achieves dynamic curvature tracking and Cartesian surface tracking and following.