The Self-Running Factory That Automates Robot Design

Modern tools and technologies are rapidly speeding up the pace of scientific discovery. From automating processes like formulating and testing thousands of compounds in the field of materials science, to analyzing diagnostic images to better optimize medical tools, scientists are turning tedious, drawn-out tasks into automated processes.

Markus Nemitz, associate professor and Jon Levy Faculty Fellow, is studying how to automate the design and fabrication of soft robotic parts. With a grant from the National Science Foundation, he and his collaborator John Rieffel at Union College are building EvoFab, a self-running “factory” that designs, 3D prints, and tests soft robotic parts. These parts are made from flexible but resilient materials, rather than the rigid metal usually associated with robots. 

“Automating how robot parts are designed and made could help the United States build more of its robots at home and strengthen domestic manufacturing, ” Nemitz said.  

The importance of soft robotics and automation

When people picture a robot, they tend to visualize a boxy, metal contraption—like something from the Disney movie “WALL-E.” In reality, this type of robot gets damaged easily. The Nemitz Robotics Group builds robots from soft components that are more resilient and can absorb these impacts. 

“When my own three children tried to take their first steps, they fell again and again without getting hurt, because their soft bodies absorbed the impact of their fall,” Nemitz explained. “We do not have anything like that in robotics. We are trying to create robots that are far more human-like to allow them to withstand the many physical impacts that so often present themselves in the real world.”  

EvoFab starts with actuators—a part of the robot that acts as the “muscle,” converting energy into physical movement. The actuators are made out of a flexible, rubber-like material that is able to bend and stretch. 

Actuators are critical to a well-functioning robot, and they are typically designed by hand; someone creates the design on a computer and simulates its functionality, and then 3D prints it to see how it behaves in the real world. The problem is that a printed actuator rarely behaves exactly like its simulation. For example, an actuator simulated to bend at a certain pressure might actually bend at noticeably more or less pressure once it is printed. “Simulating, fabricating, and testing these actuators takes an enormous amount of time and material,” Nemitz said.  

EvoFab automates this cycle. An evolutionary algorithm, developed with Rieffel’s group, generates candidate designs and screens them in a computer simulation. The most promising designs are 3D printed from soft plastic pellets. A robotic arm then removes the actuator from the print bed, inflates it, and films how it bends. These measurements are fed back to improve the simulator, so its predictions match reality more closely with each round. 

Because most of the search happens in simulation, the system only needs to print and test the strongest candidates, which saves time and material. The first version of EvoFab is now running and has autonomously printed and tested its first soft actuator; the team’s next step is to run the full design loop over many generations.

Researchers remain essential to the success of creating the parts. They set the design goals, choose and refine the materials, check that the simulator is working, and confirm that the final actuators perform as intended.

Life-saving applications of robotics

Faster, cheaper methods for designing and building robots are important – especially in scenarios in which those robots are expected to be lost. Nemitz’s long-term goal is to develop and operate robots that are capable of clearing landmines, which mostly harm civilians, many of them children. The strategy is to build low-cost (<$100) robots that are less expensive than the landmines they target. Able to navigate rough terrain and locate mines, these robots ultimately sacrifice themselves by detonating the explosive without risking human life.

“Robotics, while a very technical field, is also a very creative and interdisciplinary one,” Nemitz said. “We’re always looking for undergraduate and graduate students, with or without an engineering background, to work with us.” Learn more about the Nemitz Robotics Group. 

The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the National Science Foundation. Research reported in this article was supported by the National Science Foundation under the following award number: 2438415.