The X research group used to be known as Google X. It’s a semisecret skunkworks founded in 2010. Today, it operates as a subsidiary of Alphabet Inc. with headquarters about a mile and a half from Alphabet’s Googleplex in Mountain View, Calif.
The CEO and “captain of moonshots” at X is an entrepreneur scientist with the curious name Astro Teller. Even though the moonshot projects all begin with parallel dedications to research and development along with a mandated search for reasons why the project should be canceled, Teller has stuck with the Everyday Robot project despite its slow progress.
In a November 2021 X blog, Chief Robot Officer Hans Peter Brøndmo explains the essential problem slowing the development and what his team has decided on for an essential way forward.
“For the last several years, my team and I have been working to see if it’s possible to teach robots to perform useful tasks in the messy, unstructured spaces of our everyday lives. We imagine a world where robots work alongside us, making everyday tasks—like sorting trash, wiping tables in cafes, or tidying chairs in meeting rooms—easier.”
Everyday Robot cleaning a table. Image courtesy X Development LLC
The tasks list for the robots, he adds, won’t require much from the upper limits of AI, but will require that the robots be able to display the ordinary physical skills we all use to maneuver around and operate within our environments. And that’s the speed bump the X group is currently struggling against. What we find commonplace and easy turns out to be extraordinarily difficult for robots.
It will be worth the effort, though, Brøndmo promises because “In a more distant future, we imagine our robots helping us in a myriad of ways, like enabling older people to maintain their independence for longer. We believe that robots have the potential to have a profoundly positive impact on society and can play a role in enabling us to live healthier and more sustainable lives.”
A LOT TO LEARN
Brøndmo offers a list of typical events we take for granted. “Everyday environments like our homes or offices aren’t governed by a set of straightforward rules that robots can follow…. Even everyday objects, from chairs to coffee cups, appear, move, and disappear in ways that we expect and anticipate, but that are very mysterious to a robot. Where humans naturally combine seeing, understanding, navigating, and acting to move around and achieve their goals, robots typically need careful instruction and coding to do each of these things.”
Coding a series of maps for navigating the overwhelming variety of physical contingencies seems almost impossible. And if writing coded instructions for all the myriad circumstances was the single path taken, the project probably would have been canceled.
One of the experiments tried last year involving reinforced learning (RL), and simulation offered a way that seems to have a future. Brøndmo explains, “For robots to be useful in everyday environments we need to move away from painstakingly coding them to do specific and structured tasks in exactly the right way at exactly the right time. We have concluded that you have to teach machines to perform helpful tasks; you cannot program them.”
Then you can add an additional layer possible with the inherent AI to allow the robots to learn on their own through trial-and-error correction. In the engineer’s words, “We investigated how robots can learn from human demonstration, from shared experience, and how we can accelerate learning by simulating robots in the cloud.”
We’ve seen this before, recently with AI-enabled computers practicing overnight without supervision to learn expert skills for simple computer games like Space Invaders, the more complicated human endeavors like Chess and Go, and even the strategy of bluffing in draw poker (the program Pluribus in 2019).
Brøndmo describes the success of the experiment in the blog. “In collaboration with DeepMind and Robotics at Google, we developed a new technique which leverages simulation to reduce the amount of real-world data that RL algorithms need by more than 100 times. When applied to the task of grasping arbitrary objects from clutter using just a camera image as input, our new method is able to achieve a 70% success rate by training entirely in simulation. With just 5,000 additional grasps in the real-world, the success rate reaches 91%; a result that previously took over 500,000 grasps to achieve. Put differently, a skill that previously took us over three months to train using seven robots now takes less than a day.”
One of the goals selected for Everyday Robots is the ability to sort waste for recycling. They need to divide “caps, bottles, snack wrappers, and more across landfill, recycling, and compost bins.”
For this. the researchers added one other element to the learning techniques—collaborative learning. “Each night, tens of thousands of virtual robots practice sorting the waste in a virtual office in our cloud simulator; we then move the training to real robots to refine their sorting ability. This real-world training is then integrated back into the simulated training data and shared back with the rest of the robots so that the experience and learning of each robot is shared with them all.”
PROGRESS REPORT
The latest X blog concludes with the conviction “that reinforced learning and simulation to teach them to sort trash is possible.” And further, it reports “We are now operating a fleet of more than 100 robot prototypes that are autonomously performing a range of useful tasks around our offices. The same robot that sorts trash can now be equipped with a squeegee to wipe tables and use the same gripper that grasps cups can learn to open doors.”
The group has also decided to move out of the rapid prototyping environment of X and will focus on sending the pilots to some of Google’s Bay Area companies. And as a final sign of their optimism, the group will be dropping the “project” from their name and will now simply be known as Everyday Robots. The horizon has shifted. “For now,” Brøndmo writes, “we’re focused on teaching the robots new tasks and making sure they don’t get stuck in the corridor on their way to help us.”
