DARPA’s Learning Introspective Control (LINC) program successfully demonstrated machine learning technologies that help systems adapt to challenges encountered in the real world.
Mar 07, 2024
Imagine a world where the number of vehicle accidents is cut in half.
DARPA's Learning Introspective Control (LINC) program is developing machine learning (ML) methods that may bring that scenario closer to reality.
LINC aims to fundamentally improve the safety of mechanical systems – specifically in ground vehicles, ships, drone swarms, and robotics – using various ML-driven methods that require minimal computing power. The result is an AI-powered controller the size of a cell phone.
At Sandia National Laboratories' Robotic Vehicle Range, LINC researchers used U.S. Army robots as surrogates for larger vehicles to test their solutions, allowing the small vehicles to respond to obstacles in real-time.
"These systems use sensors while operating but have difficulty adapting when encountering unforeseen situations," said John-Francis Mergen, DARPA's LINC program manager. "Humans are very good at figuring out how to keep going when faced with a challenge, but the same cannot be said for machines. So, if we could make systems safer with enhanced controls enabled by machine learning, we'd save many people's lives."
Not only did the ML algorithms work as intended, but new behaviors also emerged as pleasant surprises to the research teams. In one instance, high winds damaged the robot's treads; however, the robot figured out how to leverage the wind and position its body as if it were a sail to help propel itself to finish the obstacle course up an incline.
Experimentation will continue in 2024 in larger systems such as light aerial multipurpose vehicles (LAMVs) and boats.