For the past several years, the popular framework for thinking about self-driving cars was the “race to autonomy,” a technological arms race with a massive pot of gold at the end of the rainbow. In the afterglow of the smartphone revolution, it’s easy to assume that on-demand robotaxis will become the kind of disruptive platform that an entire new economy could be built on with inevitably lucrative results for their creators. But, as the hype around AVs settles down it’s becoming increasingly clear that reinventing mobility is far more than just a technological challenge: in particular, the economic challenges of using incredibly expensive autonomous vehicles in Uber and Lyft’s already-unprofitable ride-hailing business are becoming increasingly clear.
In this more sober light, things that seemed weird and gimmicky a year or two ago start to make more sense. For example, if you start to think beyond the technical challenges of developing AVs and think about the business challenges they present, the “fauxtonomous” vehicles that Ford has been testing in Miami with Postmates and its autonomous pizza delivery trial with Dominos seem more valuable than they did at the time. After all, even if a delivery vehicle can drive itself to your house you still need to figure out how to get that delivery from the curb to your door, or if people are willing to come out to pick it up themselves. These are things that could be speculated on, but until you’ve done the testing you just don’t know.
Apparently these trials have surfaced real challenges with using autonomous vehicles for delivery, since Ford is taking the seemingly extreme step of recruiting a headless humanoid robot to ride along in its AVs and take deliveries right up to your doorstep. Again, this is something that can seem gimmicky and sci-fi for the sake of sci-fi, until you start to see wheeled delivery bots spinning their wheels, getting stuck in snow, getting kicked and otherwise relying on the kindness of strangers. The world that delivery bots need to navigate is designed and built for humans, giving humanoid robots a meaningful advantage over their smaller wheeled brethren.
Of course humanoid robots have their own challenges: they are larger, more complex and less efficient than a smaller wheeled robot. That’s why they make so much sense as a “co-bot” partner for an AV: rather than performing the entire delivery themselves, they simply ride along in one of Ford’s vehicles and make the quick trip from the curb to your door. Though the Agililty Robotics “Digit” that Ford is incorporating into its delivery AVs has its own camera and lidar sensors, it can also tap into the AV’s map and more powerful sensors reducing the cost of its on-board autonomy stack.
Like Ford, Agility has a lot of execution work remaining on its half of this robotic team, the company’s CTO Dr. Jonathan Hurst told me over lunch near the company’s Albany, OR headquarters. But, like Ford, Hurst also seems to understand that technology is only as good as its application. Having spent much of his career in academia, first as a doctoral candidate at Carnegie Mellon’s famed robotics department and later as a professor anchoring Oregon State’s up-and-coming robotics program, Hurst says Agility was started as much because business opportunities were starting to present themselves as anything else.
“I wanted to thoroughly understand legged motion, fundamentally and thoroughly,” Hurst says of his academic career. This involved researching the biodynamics of gaits in everything from humans to ostriches, reducing them to the most fundamental core dynamics and then translating them into robotic machines. The “spring-mass model” that bipedal gaits share gave rise to a DARPA-funded robot called ATRIAS (short for Assume The Robot Is A Sphere) that validated this approach, proving that the right dynamic design could provide inherent agility rather than having to program it into a robot after the fact. Having cracked the biodynamic code of bipedal motion, Hurst founded Agility and began building robots that could use this breakthrough to solve real problems.
Agility’s first robot, Cassie, was a dramatic step beyond ATRIAS, boasting far more sophisticated and capable legs in a lighter, more rugged package. A somewhat unsettling set of disembodied legs, Cassie won’t be delivering packages anytime soon but is being sold to research labs as a development platform. In addition to bringing in revenue and attracting interest in the company, selling Cassie as a R&D tool also provides Agility with a pipeline of talent that has experience with their unique technology. This is a trickier and more important challenge than you might think, given that only perhaps 50 people in the world have worked with this biodynamics-derived bipedal robotics technology and Agility is in the process of growing from about 20 to about 60 employees.
Digit, the robot that will be integrated into Ford’s AVs, is a direct evolution of Cassie that adds a torso and perception stack as well as arms that help it balance and recover from falls. With Digit. Hurst’s science is no longer “just a cool idea that might be useful someday,” he says. “We’ve reached the point where this technology is going to be pragmatically useful in the real world pretty soon.” Though Hurst can imagine countless applications for Agility’s robots, from smokejumping to theme park actors, the transition from research to development requires picking specific one. “There’s all kinds of ways this technology can be useful,” he says, “but with delivery we’ve identified one of many potential markets that we think is the lowest-hanging fruit and the biggest growth potential. I think we can be delivering packages in the next few years.”
This focus on delivery as a pragmatic business application of his breakthrough technology made Ford a natural partner, Hurst says, calling the attraction “mutual” rather than a case of one company pursuing the other. Ford’s approach to autonomous vehicles, with focused, small-scale deployments that are both experimental but also laser-focused on building a viable business, was the perfect opportunity for Agility while its robots provide a unique opportunity to solve the problems Ford’s early autonomous delivery trials ran into. “You can’t just hit the ground on day one deploying 10,000 robots,” he says. “With package delivery, if we can start right now, doing those deployments in more controlled environments, we can grow the technology, the market and the company together.”
Having demonstrated their vision with the Ford video released today, Agility is now working toward a real-world demonstration by the end of the year. If Ford sticks to its plan to deploy autonomous vehicles in Miami by 2021, Hurst thinks Digit could be ready to start making real-world deliveries by then. Even then, the relationship with Ford will have to be close and collaborative he says. “As we’re forced to face the realities of really doing this, we’ll be forced to try stuff we didn’t even realize we didn’t know,” he explains. “It’s important for us to work closely with Ford to figure out how to do robotic delivery really well, but this is also new for them. We both have a lot to learn.”