Watch Boston Dynamics' Terrifying Parkouring Robot Try to Learn How to Walk
Boston Dynamic’s Terminator-esque robot, Atlas, can do the job of a warehouse worker, deal with frustrating situations, and parkour better than Michael Scott. But it still struggles with a physical activity that most children learn how to do by the time they’re a year old: walk.
Every time you’ve seen Atlas in action, it’s been locked in a permanent power squat instead of fully extending its legs like humans. That’s because its bent knees are necessary to help the bot stay balanced. To get around this problem, a team of roboticists at the Institute for Human and Machine Cognition have been trying to help Atlas learn to walk naturally. IHMC research scientist, Robert Griffin, explains that this is key to mastering more complex movements, like being able to sprint across a pile of cinder blocks and quickly recover if it trips.
“The long-term vision is to make robots that are capable of equal locomotion feats as humans, so they can function as true human avatars,” he told IEEE Spectrum. “We’re trying now to design approaches that are capable of both precise footstep placement, such as when walking over a rock field with few, sparse footholds, and are robust to when this precision fails, such as really compliant terrain with lots of subtle height variations, using a single algorithm.”
In a paper presented at the International Conference on Robotics and Automation in May, Griffin and his team explain that by simply forcing Atlas to keep its legs straight it developed a stride that is shockingly similar to humans.
Without any help from the IHMC team, Atlas taught itself the toe-off motion, a crucial kind of movement that enables humans to walk. Initial real-time tests of this technique were pretty promising, Atlas was able to walk across uneven terrain and even resist a few gentle nudges. The researchers were also able to let it take 1.5-meter steps in simulation, which could enable to Atlas to gracefully glide across obstacles. But that has yet to be tested on hardware.
“With think with a few tricks, which can combine this robustness with precision to make a truly capable robot,” said Griffin.
At first glance, this would seem a significantly less terrifying breakthrough than teaching robots to bound up stairs and in theory chase us down. But of course, if you really wanted to overthrow humanity, you’d need to figure out how to walk among us without our noticing.