Elon Musk joins a crowded field of researchers trying to make humanlike robots a reality.
Episode Summary:
When Elon Musk talks, people listen. A serial entrepreneur has pulled forward the adoption of electric vehicles by at least a decade and launched several other ventures, including satellite launch and orbital Internet provider SpaceX. What’s the next frontier? According to Musk, it’s humanlike general-purpose robots, development of which will relay on Tesla’s in-house Dojo supercomputer. Is this for real?
Access all episodes of End of the Line on Engineering TV along with all of our other series.
Transcript of this week’s show:
To see any graphs, charts, graphics, images, and/or videos to which the transcript may be referring, watch the above video.
When Elon Musk talks, people listen. Or more accurately, when Elon Musk tweets, people listen. And what he talked about on the company’s annual AI day August 19 has yet again caused a stir. It’s long been known that Tesla has been developing in-house integrated circuits with an eye toward building their own supercomputer called Dojo. This is necessary, as Tesla’s self-driving strategy involves the aggregation of video data from hundreds of thousands of Tesla cars, resulting in a staggering amount of data feeding an enormously complex algorithm.
Logically, while they’re there, it in made sense to figure out what else they can do with that kind of computational horsepower. It looks like what Musk has in mind is a Star Trek style android. Now we’ve all seen industrial robots, and even some vaguely animal and humanlike advanced products from companies like Boston Dynamics, but in form they only approximate living creatures. Musk plans to introduce a very humanlike bipedal robot, about 5 foot eight and a little over hundred pounds with the ability to stand up, walk and use hands with four fingers and opposable thumbs, just like a human.
To say that this project is ambitious is an understatement. Boston Dynamics, widely regarded as the leader in this kind of robotics, has spent a decade to get bipedal robots that can move like a person and interact with their local environment. But even their Atlas relies on preloaded mapping of its overall environment, with advanced vision systems and time-of-flight sensors to infer details about things immediately in front of it. Musk is suggesting something much more lifelike, and is serious enough to announce that recruitment for coders and engineering personnel for this project has already begun.
He is also talking about building a prototype in 2022, which is likely a stretch goal. Can he do it? A key question is going to be what basic philosophy Tesla will use in sensors and data processing. Tesla has famously rejected the self-driving model of other companies in the autonomous driving space, systems that are very similar to Boston Dynamics’ approach with Atlas, mapping with sensors filling in local information.
For self-driving cars, Musk is chasing a general solution, which doesn’t require a preloaded map of the car’s general environment to drive like a human. It is reasonable to assume that his goal is to adapt that technology to the robot problem, although there will be several added problems: bipedal motion, the need to carry batteries with enough power to drive sensors, actuators and power-hungry processors, and the need to drive end effectors with the accuracy, precision and delicate force feedback necessary to do useful things. And do it all without hurting anyone. That’s a tall order.
Can he do it? Probably. Can do it by next year? Highly unlikely. But if it can be achieved in even 10 years, it will change the world far more than electric cars.