It looks like a telescoping grabber tool attached to a hat stand, but it could be the future of home robotics. Meet Stretch, the first device from Hello Robot, a startup founded by former Google director of robotics Aaron Edsinger and Georgia Tech robotics professor Charlie Kemp, that came out of stealth today after three years in development.
Stretch is not a consumer robot that’s ready to roll into your apartment and start doing the dishes, but rather a research platform that Edsinger and Kemp hope will lay the groundwork for home automation in the years to come. The bot’s lightweight and low-cost design could be a blueprint for future robots, especially those designed to help look after the elderly or people with physical disabilities, allowing companies to automate a range of household tasks the way the Roomba has automated vacuuming.
“Our long term mission is to see these types of robots in homes and workplaces being useful and helpful to people,” Edsinger tells The Verge. “What we anticipate now, though, is research labs, corporate R&D, and venture-backed startups using this for a variety of applications, all of which will move the ball forward for these mobile manipulators.”
Built to be simple
The problem with these “mobile manipulators” — as robots with grabbing arms and hands are known — is that they’re big and expensive, creating a bottleneck for researchers.
Yes, some companies do show off humanoid robots at trade shows that they claim can vacuum the floor or fetch you a beer from the fridge, but these machines are just puppets, only able to work in limited circumstances and no more functional than Nintendo’s R.O.B. robot toy. At the other end of the spectrum, mobile manipulators used by academics and researchers, like Willow Garage’s PR2, are far more capable but also heavy, complex, and pricey. The PR2, for example, weighs 220kg and costs hundreds of thousands of dollars. It’s not easy to move about or test in new environments, and it represents a sizable investment.
“It’s like a giant old computer when what you want is a laptop,” says Edsinger.
Stretch, by comparison, has been designed with simplicity and low cost in mind. Instead of using an expensive industrial arm with multiple joints and degrees of freedom, it has a single telescoping grabber attached to a central hoist that achieves a similar range of motion at a fraction of the price. A single Stretch unit costs $17,950 — still expensive when compared to the consumer market but a bargain for academics and researchers.
The arm is touch-sensitive for safety and can carry items up to 1.5kg in weight. It definitely takes a hit in terms of functionality (it’s not strong enough to open a fridge, for example), but it’s well-matched to moving things about on flat planes like tables, shelves, and counters.
Stretch has a wheeled base like a Roomba that’s 34 by 33cm in size, making it small and nimble enough to maneuver around tight spaces like corridors and small kitchens. And it uses an inexpensive depth-sensing Intel RealSense camera and 340-degree LIDAR sensor to survey and navigate its surroundings, allowing it to be operated remotely or autonomously. There’s also a microphone array built into the head that supports speech recognition for voice commands and a programmable LED light ring to give feedback to users.
If any single design choice exemplifies Hello Robot’s “simple is better” philosophy, though, it’s the robot’s grabber: a minimalist piece of hardware built from a pair of rubber cups and some metal springs.
Kemp says that when designing the bot, they quickly dismissed the idea of using human-like hands, which are overly complex and liable to break. Instead, they wanted a grabber that would stand up to real-world tests. So they looked to a ready-made research group: users on Amazon reviewing home grabbing tools for people with disabilities.
“We went through thousands of reviews, looked at the best reviewed grippers out there, used by real people in real homes grabbing objects they really wanted, and this [design] came out on top, both in the evaluations on Amazon and in our labs,” says Kemp.
“We were able to make a robotic version of it and it’s really just so versatile and forgiving. It doesn’t have to be in the exact right place to work, it’s just good at grabbing onto stuff.”
Who’s driving this thing?
While the hardware is polished, the unsolved problem for Stretch (and for home robots more generally) is how will these machines be controlled? Are they going to use AI to carry out tasks autonomously, or will they be operated remotely, like the wheeled delivery robots used to deliver groceries and takeouts?
Originally, says Edsinger, the team envisioned Stretch as a consumer product that was teleoperated: something that could be used to help people in assisted living scenarios, for example. They soon realized, though, that the economics of this didn’t work out. Stretch is relatively capable, but like all mobile manipulators, it’s incredibly slow. And if you’re going to hire someone to control a robot from afar for hours at a time just to carry out some simple jobs, it’s cheaper to get them to do the work in person in a matter of minutes.
“As we got into it, it became clear that some level of autonomy was needed,” says Edsinger.
To that end, the company has imbued Stretch with some basic autonomous capabilities. It can navigate by itself around rooms and grab and pass objects once the action has been initiated. But it can’t carry out complex tasks like folding clothes or cleaning surfaces by itself. That’ll be up to other researchers to test and implement.
The agnosticism in Stretch’s design, though, reflects the desires of Edsinger and Kemp to make sure the project has longevity. As the mantra of the robotics industry goes: robots are hard, and startups in this area fail frequently when faced with real-world challenges.
Edsinger is a veteran of Google’s abortive foray into robotics. The startup he co-founded, Meka Robotics, was bought by the search giant in 2013 along with a string of other robot companies, including Boston Dynamics. There were high hopes for Google’s entry in the market, but the company’s efforts fizzled out as it realized the limitations of consumer robotics.
Even, now, says Edsinger, “the business case isn’t quite there yet for a consumer robot ... the technology isn’t quite there yet.”
The answer, he says, is to not rush things. He and Kemp have avoided the “rocket fuel” of venture capital in the hopes of building a sustainable business through sales. The research edition of the Stretch robot has already been sold to a half-dozen research labs, which they say will provide invaluable feedback on what the market really needs.
“We realized that the best thing for us to do to really solve this problem is to be around for a while,” says Edsinger, “to not take a giant swing for the fences and disappear if we miss.”
Hide and seek, with robots
For all of the unanswered questions facing Hello Robot, Edsinger and Kemp are certainly adamant about the possibilities of Stretch. It’s just simpler and easier to use than anything else on the market, they say, and it opens up a lot of possibilities for home robotic research.
Kemp, for example, has been using Stretch prototypes around his house for years. He’s used the robots to play games with his children and carry out basic chores. One Thanksgiving, when he was away from home with family, he even used Stretch to feed his cat. Controlling the robot over the web using his laptop, he was able to open up tins of cat food and tip them into a bowl as well as fill up glasses of water from the refrigerator and pour those out.
“I’ve had this long term dream of having a versatile robot in my home, and now I’ve got one, it’s just a lot of fun,” he says. “If you have a $400,000 robot it’s really hard to imagine hiding Easter eggs with it for your kids. It constrains your thinking when you have these huge, expensive robots.” But it’s not so with a machine that costs less than $20,000.
Both men are understandably cagey when it comes to suggesting a timeframe for when a robot like Stretch might be doing useful work as a consumer device. But it’s certainly a matter of years, perhaps even decades. They point out that the Roomba (“a phenomenal product”) took decades of work to get to its current abilities, and it only does a single, relatively simple task. They’re hoping to create machines that do much, much more.