When it comes to innovation, they say two heads are better than one—but what if one of the heads is a robot?
Thanks to a $1.4 million grant from DARPA (Defense Advanced Research Projects Agency), Stevens Institute of Technology is partnering with Perspecta Labs, an expert in cyber–physical systems, and Princeton University, a leader in machine learning, to push the boundaries of how humans can put their heads together with robotic technologies to design engineering systems more quickly and more creatively.
Specifically, this trio, dubbed “AIMED” (Artificial Intelligence Managed Exploration of Designs), will be spending the next four years leveraging artificial intelligence (AI) and machine learning to enable capabilities that can help humans with complex engineering design tasks.
“Traditionally, humans have provided the inspiration for innovative designs, and the tools have provided the perspiration to automate menial tasks,” explained Kishore Pochiraju, professor in the Department of Mechanical Engineering, who is working on the grant at Stevens with Brendan Englot, associate professor of mechanical engineering. “But humans can rarely explore all possible designs, and find it particularly challenging to unearth novel, cross-disciplinary designs. Moreover, when multiple disciplines are involved, human designers are challenged by the cognitive load required to navigate complex design spaces.”
That’s where AI comes in, paving the way to completely reinvent the engineering design process.
“We’re bringing AI into the process in a more direct and meaningful way,” Englot said, “so that the role of humans can be made less burdensome and strenuous, and humans can focus on the higher-level tasks that they're the best at handling. AI can help us bridge the divide among the domain experts and innovate in ways where all the relevant disciplines can be considered simultaneously to allow new, interesting, unconventional, and even superior combinations to emerge.”
One if by air, two if by sea: Using AI to design an air taxi and a long-endurance unmanned underwater vehicle
The DARPA initiative presents the nine participating teams with two design challenges: an air taxi that can operate like a drone to transport people through urban settings, and a long-endurance unmanned underwater vehicle that can cross the ocean by itself. Each is a challenging, cutting-edge design problem that requires expertise in many disciplines.
“The aircraft is expected to take off and land vertically like a helicopter, but also transition to fixed-wing flight,” Englot said. “And there are lots of ways that an underwater vessel can harvest energy from the environment, such as the wind or the waves or the sun, and then it also has to be able to travel reliably without human intervention for long distances, collecting data and keeping track of its location. The intent is that the AI system will find unconventional, optimal combinations for both of these concepts.”
These novel designs will be inspired by some completed and successful “seed” designs, as well as what DARPA calls a “corpus”: an assemblage of existing data including similar but incomplete designs, components, goals, constraints, rules, policies, and learnings.
“The AI system will be tasked with reaching across disciplines to propose optimal designs, and then the humans will provide feedback on whether the designs could be realistic and feasible,” Englot explained. “That's why we call it AI Managed Exploration of Designs: the AI is managing this process where it's learning from all the data that's been presented and making suggestions, and the humans can make high-level choices and provide high-level feedback. This has potential to speed up the innovation and design process, so that rolling out something like a new aircraft model or military vehicle no longer takes years to accomplish.”
Some of the DARPA teams are investigating how humans will interact with the system, and others are designing the challenge problems. Stevens is focusing on the designs emerging from the symbiotic process to find the novel solutions. The AIMED team is leading the development and integration of high-fidelity simulation tools for domains including controls, structures, materials, fluid dynamics, and energy, and the creation of digital twins—that is, virtual replications of the designs that help predict their viability.
“Stevens has a lot of expertise in design, manufacturing, and also all the disciplines required to create a digital twin,” Englot said. “We're acquainted with all the modeling and simulation tools that we'll need to simulate these designs.”
Putting the design systems to the test
Throughout the project, DARPA will periodically host hackathons in which the teams will test and qualify their tools against design challenges.
“They want us to have our systems reliable enough that, on short notice, we could tackle and propose solutions to any problem,” Englot noted. “The first hackathon is coming around April, and we want to have a system ready to test on the first challenge problem.”
There will also be a competition in which engineers using traditional design approaches compete against the AI systems.
“It’s unique that we’re able to go beyond proof of concept to do something much more meaningful and build a much more complex set of tools because we have such a large multidisciplinary team,” Englot said. “As faculty, it’s not often that we get to work with such large teams, and have such ambitious goals, which makes this project even more exciting.”
Learn more about mechanical engineering at Stevens: