Robotic arms in the NERVE Center can perform a variety of tasks. | Source: UMass Lowell, Brooke Coupal
Everyday tasks for humans, like simply picking up a water bottle, are still a huge challenge for robots. Robotics researchers and developers have been working for decades to enhance robot manipulation skills, but there’s a broader issue slowing their work and making it difficult to innovate.
This issue is a “full system problem,” according to Adam Norton, the associate director of the University of Massachusetts (UMass) Lowell New England Robotics Validation and Experimentation (NERVE) Center. For a robot to complete a simple task, many parts have to be considered – from the gripper used to grasp an object to the software that controls the robot. Researchers who want to develop one aspect of the manipulation process also need to figure out a way to build all of the other pieces of the robot before testing their development.
While research exists to help mitigate these issues, the diversity of this work – from the technology researchers use to their methods – makes it difficult for others to incorporate it into their research.
“It significantly holds back researchers who have to do all this work before actually implementing the part that is their contribution to the field,” Norton said. “It creates an enormous barrier to entry.”
This is why Norton and Holly Yanco, chair of the Miner School of Computer & Information Sciences in the Kennedy College of Sciences and director of NERVE, along with a team of researchers, are looking to improve robot manipulation by developing what they call a standardized “ecosystem.”
Adam Norton (left) and Holly Yanco, who lead the COMPARE project. | Source: UMass Lowell, Brooke Coupal
The researchers say this ecosystem would be a community-driven repository of guidelines, activities, and working groups that will allow for open-source, or publicly available, research to be easily compared and implemented.
“By establishing a new community-driven, open-source ecosystem for robot manipulation research and benchmarking, the field will move more rapidly toward solutions for robot perception and grasping,” Yanco said.
The program is called the Collaborative Open-source Manipulation Performance Assessment for Robotics Enhancement (COMPARE). COMPARE is being funded by a National Science Foundation (NSF) grant worth nearly $1.5 million.
“Everybody working in this space is approaching it in very different ways,” Yanco said. “The goal of the COMPARE ecosystem is to create greater cohesion and compatibility between the research efforts.”
Yanco and her team plan to work with the robot manipulation community to develop research standards that will establish cohesion between the work of roboticists. The team believes that by following the same standards, researchers can compare research shared in the COMPARE system with their own work. This gives them the opportunity to implement the research of others into their efforts.
Funded by a previous NSF grant worth nearly $300,000, Yanco and Norton ran workshops with robot manipulation researchers. These workshops gave them the opportunity to learn what researchers wanted in a standardized system.
Since these workshops, Yanco and Norton have created an advisory board featuring people from the industry, government organizations, and academic research institutions for continued feedback. The team is also working with robotics experts at Rutgers University, Yale University, University of South Florida, and Worcester Polytechnic Institute. All these institutions have different specialties that can contribute to the system.
“By bringing people in, we’re hoping they will go off into industry or academia and use these standards as part of their research,” Yanco said. “This could have a big impact on robot manipulation.”
As guidelines are developed, more workshops will run at the NERVE Center and other robotic facilities, where researchers can put the standards into practice.
The post Making it easier to compare robot manipulation research appeared first on The Robot Report.
