Robotics company Boston Dynamics has a new four-legged addition to its family: a 160-pound quadruped named Spot—and dude is impressive. In a short video posted yesterday, the Google-owned company shows Spot gallivanting with a weird equestrian hop through office corridors, up concrete stairs, and along rocky hills with little trouble. Spot even keeps his balance after being kicked by one of its human overlords. You can almost hear the canine whimper over the drone of his hydraulics.
We’ve seen all of this—admittedly amazing—stuff out of BD’s four-legged robots before. But it gets crazier around the 1:20 mark, when a pair of Spots begin trekking up a hill. Spot Number One starts repeatedly colliding into Spot Number Two—and neither loses balance. After a few seconds and a bit of subtle push-and-shove, they straighten out and walk in parallel again, and then turn together once they reach the top of the hill. This is getting creepy, guys—it looks like these robots are exhibiting the same swarm-like behavior that we see in animals.
We checked in with Iain Couzin, a Princeton biologist and expert in the study of collective animal behavior, to get his take on the robots’ seeming hive mind.
We know from Spot’s reaction to that kick that he can dynamically correct his stability—behavior that’s modeled after biological systems. From what Couzin can tell, the robots’ collective movement is an organic outgrowth of that self-correction. When the two Spots collide at the 1:25 mark, they’re both able to recover quickly from the nudge and continue on their route up the hill. “But the collision does result in them tending to align with one another (since each pushes against the other),” Couzin wrote in an email. “That can be an important factor: Simple collisions among individuals can result in collective motion.”
In Couzin’s research on locusts, for example, the insects form plagues that move together by just barely avoiding collisions. “Recently, avoidance has also been shown to allow the humble fruit fly to make effective collective decisions,” he wrote.
It doesn’t look like Spot is programmed to work with his twin brothers and sisters—but that doesn’t matter if their coordination emerges naturally from the physical rules that govern each individual robot. Clearly, bumping into each other isn’t the safest or most efficient way to get your robot army to march in lock step, but it’s a good start. And it’s relatively easy to imagine several Spots working together in organized ways if the LIDAR sensors fitted on their “heads” were programmed to create avoidance behaviors—like those locusts—rather than simply reacting to collisions.
Spot’s life-like motions are uncanny, but when you add this emergent, collective behavior—which can sometimes be unpredictable—the possibilities get downright scary. Will Spot’s group dynamics stop at the point of swarming like locusts? (Ominous.) Will they cluster into self-protecting balls like sardines? (Less so.) Or could they end up as smart and responsive as humans?
Couzin goes so far as to call this bump-and-grind between Spots One and Two a social interaction. “No matter how primitive, there’s no doubt that these interactions could enhance the decision-making capabilities of such robots when they must make their own, autonomous, decisions in an uncertain world,” he wrote. We’ll just have to hope that decision-making involves not trampling us when a pack of Spots starts stampeding like wildebeest.
No comments:
Post a Comment