From Shinkansen bullet trains to suits for swimming, nature has always been an engineer’s greatest source of inspiration. Now, researchers are turning to the human eye as they look to accelerate physical AI, the branch of artificial intelligence that lets machines interact with and adapt to the physical world. In an interview with The News-Letter, Charbel Rizk, a former professor of Electrical and Computer Engineering at the Whiting School of Engineering, shared the work that his startup, Oculi, is doing in this field.
Rizk, who hails from Jezzine, a mountainous town in Lebanon (a country he emphasized is, in fact, not a desert), became interested in autonomous vehicles while pursuing his PhD at Georgia Tech.
“[I participated in] a competition for autonomous drones to fly from point A to point B, pick up an object, deliver it to point C and come back to point A.”
His team became the first to conceptualize and build a quadcopter. This four-rotor design is very stable because two rotors spin clockwise and the other two spin counterclockwise, generating equal and opposite torques that keep it in the air. Rizk recounted demonstrating the design to a U.S. Navy captain, an encounter which suggested people can often not see immediate value in ingenious ideas.
“[The captain] looked at it and said ‘That’s a pretty cool toy, but what do I do with it?’”
However, they could only afford to carry one camera on board, given the bulky size of the computer that processed video in real time to aid in navigation.
“That’s really when I realized that the world of AI, the world of autonomy, is not going to take off unless you solve the vision problem.” Rizk said, alluding to his eureka moment.
Rizk corroborated the hypothesis that vision was the limiting factor in translating many technologies into real-world products while working in the aerospace industry and collaborating with the defense world, settings in which people still relied on the traditional camera and computer set-up to collect data and make decisions in real time.
Inspired by the human eye’s ability to process visual information alongside the brain, detecting contrast, color and minute details, Rizk founded Oculi. Its technology centers on the sensing and processing unit (SPU) it created. The SPU is composed of pixels with individual logic and memory, allowing each pixel to process the visual data it captures and determine whether it is of interest before sending it to a downstream processing unit. This approach can minimize energy consumption by reducing the power required for data transfer.
Additionally, Oculi offers an enticing solution for data privacy, as its technology can make decisions without even generating an image of the scene. However, a trade-off remains between collecting more data — on the off-chance it is needed — and giving up data that might be useful.
“You can save up to $50 billion in energy every year if you make them smart,” Rizk described.
Oculi’s main focus is commercial applications of its SPU. Take sliding doors, for instance. Rizk believes that instead of going after superintelligence, current work in the field should stop and consider basic functions in the world that still aren’t well automated, like why a sliding door opens even when walking past it or just waving a hand in front of it. With Oculi’s SPU, this issue could be tackled by programming the door to open only when someone walks toward it with the intention of passing through.
Rizk explained these applications might not have the same allure as using an SPU in unmanned defense vehicles operating in off-grid terrain, but they could have an immediate impact. In Northern Japan, for example, black bears sometimes venture into urban environments in winter to find food. There have been instances where bears have gone into supermarkets simply because a sliding door opened for them, a calamity that could easily have been avoided with an SPU. Adding vision can also reduce pedestrian deaths, help the visually impaired community and cut energy and money wasted on heating, cooling and lighting.
Rizk is fueled by problems like these, which exemplify what it means to engineer solutions for humanity’s greater good. He is quick to point out, however, that in startups — unlike in academia — he has to consider not just what he wants to do, but also what his investors want. When asked what advice he would give students grappling with different career options, he was clear: be honest with yourself and ask whether you’re competent and self-motivated.
“[If you are,] you should go start your own company or join a startup or small company,” Rizk remarked.
The future for physical AI is promising, particularly as startups like Oculi pave the way for dynamic solutions that can revolutionize everything from traffic management systems to electronic devices. Rizk firmly believes the vision problem is the Achilles’ heel we need to tackle to engineer a safer, greener and more efficient future.
