Safety researchers develop systems to make advanced manufacturing in New England safer

A coalition of experts and institutions from across New England have been working for two years to expand the use of intelligent manufacturing. The Northeast Integrated Intelligent Manufacturing Lab (NIIM) was established through a collaborative NSF EPSCoR Research Infrastructure Improvement Program: Focused EPSCoR Collaborations (formerly known as Track-2) Program award in 2022.
The NIIM lab, based at the University of Maine, is focused on intelligent manufacturing, which is becoming increasingly embraced. “Modern manufacturing is moving a lot towards 3D printing and human collaborative robotics,” remarked Vikas Dhiman, an assistant professor of computer engineering working on the NIIM project. Robots already work in close proximity to humans in this modern age of manufacturing, which provides huge benefits to productivity. The NIIM project envisions advancements in robotics and manufacturing that would benefit the Northeast, especially through the addition of artificial intelligence (AI) and machine learning.

In manufacturing, the use of robots has been limited to precise and highly specific tasks, such as “drill a 2mm hole at a 2cm location in this plate.” Recent advancements in AI have changed the scope of tasks, and created the possibility of robots working on loosely defined tasks, such as “sorting out defective parts.” Moreover, AI allows the robots to work in the same physical space as humans, instead of being confined to a cage. However, collaboration between humans and machines creates new safety concerns. The NIIM project has made safety a top priority. Researchers are studying how AI integration can help improve human safety when working in close proximity with robots. For instance, robots need to use cameras and sensory equipment to detect humans and operate around them.

Safety researchers using AI are presented with a dilemma. A robot needs to track a person at all times to prevent a collision and worse. In this field, accuracy is critical. “AI is very good at improving accuracy, but it also brings uncertainty because all AI algorithms are interpolation algorithms,” said Dhiman. “One can think of interpolation algorithms as connecting dots. Any given set of dots can be connected in multiple ways. This choice leads to uncertainty which is inherent in all AI algorithms.” Currently, the open research questions focus on estimating and reducing uncertainty, as well as making decisions that are aware of this uncertainty. The safety application of AI has become a broader discussion in the robotics field.

Dhiman is studying this uncertainty in AI predictions. There are several sources of uncertainty, which are often organized into two categories: aleatoric and epistemic. Aleatoric uncertainty, from the Latin alea for die, concerns random chance. Epistemic uncertainty, from the Greek episteme for knowledge, concerns uncertainty that comes from the lack of data. Dhiman noted that the NIIM team would have to estimate both. “It is important to account for both kinds of uncertainty, even if it requires separate algorithms.” To quantify these sources of uncertainties, the NIIM team has been using a multitude of methods, including developing Bayesian neural networks. Bayesian algorithms start with assumptions and collect data to update those assumptions, not unlike how humans make predictions. More importantly, Bayesian algorithms are uncertainty-aware; they know what they do not know. These algorithms will help robots dynamically identify key parts of their environment and adapt accordingly.

In its first two years, the NIIM lab has made an impact on manufacturing research. The project has emphasized the potential uses of AI algorithms and the ways that manufacturing can benefit from embracing robotics. Dhiman hopes to develop AI algorithms that enable robots to work in close proximity with humans while probably guaranteeing everyone’s safety.