The second-generation Atlas robot sets a high watermark for humanoid robotics

Anyone who tuned in for theDARPA robotics challenge last year, covered in unsparing detail by ESPN, would likely have come away fantastically disappointed. Most of the robots could barely manage to walk upright, much less climb a flight of stairs or traverse uneven terrain. Many forms of artificial intelligence — defined as software that can mimic human-like abilities, such as Google’s speech recognitions algorithm — have made jaw dropping advancements in the last few years. But the field of bipedal robotics has not kept pace. This makes the videos demonstrating the release of Boston Dynamics Atlas 2 robot last week all the more impressive.

Boston Dynamics has gotten used to leading the way in humanoid robotics. Ever since the company was spun out of MIT in 1992, they have regularly made headlines by airing YouTube videos of their latest robotic sensations. Their next generation Atlas robot is no exception. In the accompanying video, it is shown legging over difficult, snowy terrain, and recovering adroitly after a patch of ground gives way beneath it. Hitherto, these were the kind of feats roboticists could merely dream of.

To many of us, this feels surprising. Given that robots routinely impress us with their ability to beat chess champions and embarrass jeopardy masters, accomplishments few humans will ever attain, why should walking across a snowy landscape prove so difficult?

The answer to this paradox is enlightening and lies deep in our evolutionary past. Walking over uneven terrain is a problem that evolution has been relentlessly experimenting with for the last 500 million years, with the result that we now possess some of nature’s most ingenious mechanisms for solving this dilemma hardwired straight into our genomes. You can bet that if chess skill was required for reproductive success over the last half billion years, nature would have devised some unimaginably good chess playing creatures that put even IBM’s Deep Blue to shame.

The truth is, evolution has only recently started selecting for reproductive fitness on the basis of chess playing skill. Bipedal walking, which comes easy to us thanks to evolution, is actually an incredibly challenging problem, requiring far more computational acrobatics than that presented by a game of chess.

Returning to the new Atlas robot, we see that the folks at Boston Dynamics have in fact recapitulated close to 500 million years’ worth of evolutionary experimentation in a few short decades. Compare this with IBM’s Deep Blue, which by evolutionary standards is only as old as humans have been playing chess, maybe 1,500 years. So while the reception of the second-generation Atlas robot will occasion less fanfare than the ousting of the world’s foremost chess champion by a computer, it is the former that should really give us pause. If the folks at Boston Dynamics have attained what it took evolution nearly 500 million years to accomplish, one can only speculate how far they will push the bar in the next few decades.