By Fan Lou
Imagine a robot that is made with no rigid parts. That is what Marc Kilpack, an assistant professor from Brigham Young University, demonstrated in his presentation on soft robotics. His seminar, titled “Controlling and Modeling Large Scale Soft Pneumatic Robots,” was held March 13, 2020, in the Mechanics Seminar at University of Wisconsin-Madison. Killpack is a leading innovator in the field of robotics. He and his team of graduate students and researchers in the Robotics & Dynamics Lab are at the forefront of soft robotics research.
For those who are not familiar, you might be thinking, what is a soft robot? Killpack, in his presentation explained, “A soft robot has very few to no rigid parts, it is built from soft materials like polyester and plastics.” These robots are soft, light weight, impact resistant, and have extremely high strength to weight ratio. They are fit for various kinds of tasks in domestic, health care, rescue, and exploration applications. Professor Killpack demonstrated their pneumatic arm robot mounted on a NASA rover drone completing various tasks, even while subjected to strong interferences and rough treatments that would most likely render a traditional mechanical arm to failure. Thus, these soft robots possess huge advantages in certain situations over traditional robots. Killpack and his team are helping people to understand their utility and see the benefits they can bring while pushing the engineering forefront in soft robotics control.
Killpack also discussed the current limits and some inherent weaknesses soft robots compared to conventional rigid body robots. Soft robots have a large degree of uncertainties and inherently low accuracy. Due to the pneumatic control and soft body, it is hard to control the exact motion or movement of the robot. Even if the pressure of the air that powers the robot is strictly controlled, there are many other factors to control that prevent the exact same result with every trial. They are also difficult to sense, making it hard for the controller and computer to understand the state of its body. Soft robots have many degrees of freedom compared to traditional robots with rigid bodies, thus soft robot sensing still remains a major topic of research. These issues make soft robots very hard to model and predict for conventional controllers, thus there is work remaining to make soft robots more reliable and precise.
But Killpack has some possible answers. His talk described how his research approaches modeling complex systems with machine learning and simplifying the complexities down to a processable scale for the robot on-board computer. When they use computer graphics cards to perform intense calculations they can achieve accuracy in soft robot movements similar to that of a traditional robot.
Killpack’s presentation was awe inspiring and showed that soft robots are capable of some hard achievements.