by Hans Peter Brøndmo

Oct 12, 2017

introducing-everyday-robots

This post originally appeared on the X blog.

Machine Learning + Robots = new approaches to humanity’s big problems

Pop culture created our love affair with robots; thanks to movies, TV and media going back to the 1950s and 1960s, millions of us are waiting for our own friendly bipedal humanoid. Perhaps you think that only when our laundry is automatically folded and dishwasher loaded, “the future” will have arrived. But these strong pop culture notions of what a robot is have had an unintended side effect: we often misunderstand what robots really are. “Building cool robot technology” is not an end in itself; instead, robots are tools that we can put to work to extend humanity’s capabilities.

A robotic arm setup in one of X’s robot training labs

A robotic arm setup in one of X’s robot training labs

Robots are a tool for solving some of humanity’s big problems

They may not be the robots you were looking for, but we’re already surrounded by devices that have the three fundamental components of a robot: sensing, computation, and actuation. A robot has to be aware of the world, process the information it has about the world, and then act in the world. A toaster is a very basic robot. You can buy robots to vacuum your rug and clean your pool. Waymo self-driving carsWing delivery drones, and even Makani energy kites are all robots. What they have in common is that we don’t primarily think of them as robots; we think of them in terms of the problem they can solve for us.

At X we have a number of moonshots with robotics at their core — including the one I’m fortunate to captain. Our focus is on solving problems that we believe can have a positive impact on many millions of people. In late 2015, X welcomed a number of robotics start-up teams that had been acquired by Google, and we have spent a lot of time evaluating, validating and investing in the amazing technologies they’d built. Our goal: finding paths to products that would have clear customers and real-world application.

Yet robotics still has many fundamental engineering challenges to solve, so we’ve been taking it a lot further. In close collaboration with Google Brain we’ve been combining the magic from our acquisitions with some of our cutting edge technologies, most notably machine learning. And while we have a ton of fun playing around with cool technology, our main focus is on the problems we’re trying to solve, while remaining open to the best technology to get us there.

Machine Learning + robotics gets us closer to the moon

The combination of Machine learning (ML) and robotics is one of the reasons I’m most excited to be at X today. ML will be necessary to make machines that are useful and reliable in unstructured environments. It’s at this point that robotics will really begin to open up new frontiers for attacking previously intractable problems.

X is one of the few places in the world where world-class hardware engineers, world-class software engineers and world class ML minds are working side by side, with access to virtually unlimited compute resources. And we’re home to a one-of-a-kind lab of manipulation arms purpose-built for machine learning research.

Large scale data collection with an array of robots

We’re working with the Google Brain team to explore how to teach robots new skills by learning from their shared experiencelearning from human demonstration and we’re even simulating robots in the cloud so they can train fast, and then we’re transferring this learning onto real robots. By having virtual robots we can gather lots of data for training in the cloud. Then we transfer what the virtual robots learn to the real-world robots so they can quickly learn to perform new tasks. This is all critical research that will pave the (long) path toward building machines that can learn new skills quickly and operate safely and cost effectively in the world we live in.

Using Simulation and Domain Adaptation to Improve Efficiency of Deep Robotic Grasping

If you’d like a quick primer on the current state of robotics and machine learning and a sense of the road ahead, Rodney Brooks has a useful series of blog posts.

Moving from research to the real world: the future is closer than you may think

For many applications of robotics, the “final” answer could be decades away. Robotics platforms today aren’t even in the Apple II, let alone the smartphone era. They are mostly special purpose machines that can only be programmed by experts. In the coming years I predict you’ll see many organizations — including us — try many different approaches. For example, in a field like delivery drones, you can already see some companies building highly specialized products for say, medical deliveries, while others are betting on hardware or software that could be used for a wide range of missions, from real estate photography to agriculture surveying. As always, the only way to make progress is to get building and get into the real world as soon as possible to learn what works and what people will want.

While some people may be disappointed that their dream bipedal humanoid robot future hasn’t yet arrived, others might be relieved. Those of us at X are simply excited to get to work every day on things that could make millions of people’s lives easier and turn into successful businesses. And while we’re not prepared to share the details of what we’re up to quite yet, there’s a lot of really exciting work going on as we build a bridge from science fiction to solving high-impact problems in the real world.

If you’re passionate about applying robotics to some of the planet’s most intractable problems perhaps you should consider joining us; check out our open roles.