search slide
search slide
pages bottom

Scientists isolate the brain’s ‘physics engine,’ could lead to faster, more agile robots

Physics has a reputation as a tough upper-level science course, even at the high school level. It’s therefore somewhat ironic that every human on Earth performs thousands of physics calculations on a daily basis, often without ever realizing they’ve done so. Now, for the first time, Johns Hopkins scientists have located the part of the brain responsible for this basic, fundamental processing.

It lies at the intersection of multiple functional systems in the cortex, that crucial few millimeters responsible for so much of what humans do. Principally, the hard physics calculations happen in the same cortical parcels we use to do motor planning. But there are network call-outs to the somatosensory cortex too, which integrates the acquisitive senses with the body’s own proprioceptive sense of where it is in space. These brain regions are also part of the multiple demand network, a widespread set of functional areas.

“The multiple demand center is sort of the brain’s CPU for handling generalized, difficult tasks,” Dr. Jason Fischer, lead author, told us.

In game, the physics engine has to be informed by real-world constraints like walls. In the brain, that information comes from the hippocampus and the sensory association areas. Like invoking a library, the brain’s physics system can call out to the hippocampus so it knows what the map looks like, and to the association cortex which populates the map with understandable things. Then the multiple demand network lights up. At a finer scale within the multiple demand network, said Dr. Fischer, there are regions that do motor planning, and even deciding how to use a novel tool for the first time.

Under the hood, your brain is basically made of calculus and physics, which makes this game engine analogy work on several levels. Not only does it have to perform dozens of differential equations in real-time to move you accurately through the force diagram as you move through space, the brain is itself composed of neurons whose function can be neatly modeled using differential equations. That’s handy, because it provides a useful common property that lets us extend our understanding of the brain as it does biological computation.

But what really sets the brain apart from robots is its speed. That’s why this study matters to robotics. The team expects these findings to be useful in making robots faster and more agile. BigDog and others of its kind are interesting, to be sure, but they’re almost more proof of concept than they are real-world applications, and they land squarely in the Uncanny Valley. Rather than trying to make them move naturally, though, robots that know how humans think about physics and motor planning could be incredibly useful in fields from neurosurgery to sports, and even search and rescue.

“Humans have this ability to mentally simulate the physical dynamics of the world, and to act predictively based on those simulations,” Dr. Fischer said. “For the same problems we might solve using differential equations, the brain may be implementing an algorithm that’s much simpler, elegantly simple.”

Leave a Reply

Captcha image