Recently, Bloomberg reported that Amazon has plans to delve into home robotics. Mega-tech corporations trying their hand at robotics is not a new phenomenon: Bloomberg themselves did an in-depth piece on the state of robotics at Alphabet just a few months ago.
Immediately, let’s keep in mind is that a “plan to build home robots” does not mean that the robots will actually be sold. Just as Apple was supposedly working on a self-driving car project. Not all projects survive feasibility testing and make it to the product stage. My understanding is that in 2011 or so, when Google when on their first robotics hiring spree, they attempted to make a home robot, but decided that the technology had not advanced far enough yet for it to be useful in the home. Even with the incredible recent advances in machine learning and robotics, I still don’t think we’re there yet. I’m hoping both projects succeed, because of what those successes would mean for the robotics industry in general. But Alphabet and Amazon will face resistance on multiple fronts and the deck is stacked against them. Let’s talk about why. In a phrase:
Robotics is hard.
Building a new robot is a truly difficult problem. As a robotics student, I’m not trying to claim intellectual superiority here; rather, I want to point out that building a good robot requires experts from a wide variety of fields. If algorithm performance is the 100-yard dash, and cybersecurity is the javelin throw, then building a robot is the decathlon.
A successful home robot must perform autonomous navigation, visual object detection, motion planning, and obstacle avoidance. None of these problems are new, and while the research community is making headway each year, none of those problems are fully solved yet. Welcome to robotics research. Navigation is one of the more “solved” problems for robotics, although even it is far from perfect. The question is not whether or not Amazon can solve 3D computer vision or Alphabet can robot arm motion planning. The question whether or not a robot can peform well enough to complete some useful tasks in the home by using a collection of solutions to each subproblem, all working in concert.
Unfortunately, so far we’ve only discussed the robotics-specific part of home robotics. From the electrical side, add the issues of battery life, thermal management, and safe emergency stop procedures. On the mechanical side, strength-to-weight ratios, drivetrains, and disturbance-tolerant control systems are some important considerations. And of course, robots run software, so the standard complexities of bandwith limitations and efficient algorithms also apply.
No problem I’ve mentioned so far is unique to the home part of home robotics. But what about privacy and security issues? You think Alexa listening to you snore is a violation of your privacy? When you buy an Amazon Vesta or an Alphabet Replicant, you will be putting a mobile, hackable device with a camera and an arm in your living space. It could punch in the code to disable your security system. It could unlock your door from the inside. It could drive to your desk and take pictures of your tax returns. In fact, you might even ask it to do some of these things! How do we prevent bad actors from doing the same? Will companies mine our house for personal information as well as our online behavior. This adds yet another dimension to consider when designing a robot and its capabilities. Finally, but perhaps most importantly, there are basic safety issues such as detecting collisions with humans (still a difficult problem, unless you are willing to spend exorbitant amounts on hardware) and fault tolerance.
Alphabet and Amazon both are capable of conducting world-class research that can find workable solutions to some or most of the critical robotics problems. They have hired key robotics talent away from top universities and national laboratories, bolstering their teams in areas such as human-robot interaction and deep learning. If I had to guess, I would place Google ahead on the AI side of things, due to their amazing teams at Brain and DeepMind, and Amazon ahead on the hardware problems, due to their sponsoring of the Amazon Picking Challenge and purchasing of Kiva Systems (robots that move around shelves in the Amazon warehouses). But the challenges of deploying a home robot is greater than what can be solved by any one team, lab, company, or pair of companies. Home robotics is consists of the intersection of hardware, software, and human factors–a massive hyperspace with very few habitable regions. I look forward to getting there someday, but for now, all of those regions and their respective projects are still out of our collective technological reach.