Everyone knows we don't always start with the body we want. Some people come to believe they might do better with a different brain — or at least one more attuned to the needs of contemporary society.
The same goes for robots, or at least how we feel about them. But unlike us, they may soon have "brains" far more appropriate for the tasks we have them execute, according to a recent blog post shared on MIT's official website.
New customized 'brains' for robots under development
Robots today can move very, very fast. "The motors are fast, and they're powerful," said Sabrina Neuman in the blog post.
An MIT scientist is about to present a new method — called robomorphic computing — capable of curating a robot's physical layout and design applications to generate a customized computer chip, with the aim of minimizing a robot's response time.
However, in complex environments like engaging real humans, robots can have trouble keeping up. "The hang-up is what's going on in the robot's head," added Neuman.
Tweaking robot hardware could reduce risk for hospital patients
The process of perceiving stimuli and calculating an adequate response requires a "boatload of computation," which puts a lower limit on reaction time, according to Neuman — who is a recent graduate with doctoral studies from MIT Computer Science and Artificial Intelligence Laboratory (CSAIL).
Neuman discovered a new way to surmount the mismatch between the "mind" and body of a robot. Called robomorphic computing, the new method has the potential to advance several applications in the field of robotics — including frontline medical care of patients infected with contagious illnesses.
"It would be fantastic if we could have robots that could help reduce risk for patients and hospital workers," said Neuman in the blog post. Neuman will present their findings at the International Conference on Architectural Support for Programming Languages and Operating Systems, in April.
Smooth robot operation depends on three steps
Co-authors of her work at MIT include the Edwin Sibley Webster Professor of Electrical Engineering Srini Devadas — who is also Neuman's doctoral advisor — and a fellow graduate student named Thomas Bourgeat. Other co-authors include Thierry Tambe, Janapa Reddi, and Brian Plancher of Harvard University.
As of writing, Neuman is a postdoctoral NSF Computing Innovation Fellow at Harvard's School of Engineering and Applied Sciences.
For a robot to operate smoothly, it needs to execute three main steps, in Neuman's mind. First comes perception — which involves gathering data via sensors, cameras, or other sensing instruments. Second comes mapping and the robot's ability to situate itself in the area: "Based on what they've seen, they have to construct a map of the world around them and then localize themselves within the map."
Instead of software, Neuman looks to improve hardware
The third step for the robot is to control and plan motion — creating a course of robotic action.
Taken together, these steps can eat a lot of time and computing power. "For robots to be deployed into the field and safely operate in dynamic environments around humans, they need to be able to think and react very quickly," said Plancher in the blog post. "Current algorithms cannot be run on current CPU hardware fast enough."
Researchers have sought better algorithms to circumvent this issue, added Neuman, but she thinks improving software alone won't open the door. "What's relatively new is the idea that you might also explore better hardware," she said.
Future robots could learn to tango with humans
This means hardware acceleration might enhance the performance of a standard CPU processing chip — which typically comprises the brain of a robot.
A month ago we saw robots from Boston Dynamics dance like never before (for the holidays). While impressive, the moves were obviously choreographed. But the frontier work of Neuman and her colleagues at MIT and Harvard University could usher a new age of robots capable of trading impromptu tricks with humans in a dance circle — and much more.