Q.U.R. (short story): Form follows function in robot design

Recommendation: Find a copy of Q.U.R. and read it! It’s an accurate look at designing an economical, green robot. And you’ll never look at the DARPA Robotics Challenge the same way again

Robots: humanoids, non-anthropomorphic robots

What it gets right about robots: Design and lifecycle maintenance cost

Q.U.R. is a 1943 short story by Anthony Boucher. It is an old-school, Bringing Up Baby or His Girl Friday screwball comedy, technically a bromance since the protagonists are male, set in a robotics startup company. While the comedy is a bit forced, the story accurately summarizes arguments for why morphology (form) is important in robot design and why Louis Sullivan’s 1896 dictum that form follows function still resonates today. (Well, the quote was actually "form ever follows function" but the “ever" has been lost).

The story is set in a near future where robots are commonplace, but all are of the robots are androids built by Robinc (the nickname of Robots Incorporated). The scientific MacGuffin is that, instead of building expensive general purpose android robots, the startup company will build inexpensive robots that are optimized for a single purpose. Instead of Rossum’s Universal Robots (from the play R.U.R. by Karel Capek that created the term robot), under hero scientist Doug Quinby’s direction, they will build Quinby’s Usuform Robots (“usuform” being the word they created to capture “use-oriented form" instead of “universal"— get it?). Much drinking and scientific hijinks ensue.

Robinc in Q.U.R. builds only humanoid robots for two reasons. One is because of the assumption that the human shape is the best general-purpose shape for tasks. The second is because they believe that people will only accept humanoid robots. Whether the humanoid form is ultimate general-purpose morphology is debatable in real life, but, even if it is true, it is usually more cost-effective to design a robot for a specific task. People may like cute little humanoid robots like Asimo and Pepper, but that doesn’t mean they will buy them or that they’d use them for real work, like manufacturing or automating a warehouse.

A great example of how a humanoid is not a good design for a robot is the popular iRobot Roomba. The Roomba doesn’t look like a janitor or maid who uses a vacuum cleaner and it doesn’t use the same cleaning pattern a human would. There were two major design breakthroughs in the Roomba. The first was integrating the “pusher” and the vacuum into a single, specialized unit rather than building a robot to use a regular vacuum cleaner- or to even look like a regular vacuum cleaner. The round shape helps prevent getting stuck, as does the lower height. The Roomba did not meet anyone’s expectations of what a vacuum cleaner should look like. The second design breakthrough was that the Roomba wanders around randomly, cleaning the floor the way a sheep might graze a paddock. A random wander pattern takes a lot of time to cover an area but it will cover it eventually, and if you are turning on the robot when you go to work in the morning, it has over 8 hours to get the job done. Sure, the owner could have done it in 10 minutes using a more efficient planned path around the room and obstacles, but that is irrelevant. The real design specification is “done by the time the owner comes back”, not “do it the way the owner would do it.” Starting in 2002, people paid good money for a Roomba, showing that were willing to set aside their preconceived notions of Josie, the Jetson’s robot maid, for a robot that, although it looked and acted oddly and could only perform one task, was relatively inexpensive and worked well enough. This is exactly what Boucher was getting at back in 1943.

The Robinc assumption that androids were the best shape permeates robotics research to this day. For example, the DARPA Robotics Challenge strongly encouraged humanoid robots and even furnished competitors with the Atlas humanoid robot. The tasks were aspects of enabling a robot to handle a Fukushima Daiichi type of disaster, where a robot would need to enter a process safety facility (the term for nuclear plants, chemical plants, and drug manufacturing facilities with toxic byproducts) that were still intact enough for a person to enter but too unsafe (as in being bathed in nuclear radiation or a Bhopal-like chemical release). The robot was supposed to get into a car, drive to the facility, open doors, go up stairs, climb a ladder, turn some valves, and cut a hole in a wall. The idea was that since a process safety facility was designed for humans, then only a human shape would be appropriate to try to fix a problem.

When the competition came out, my colleagues in the process safety industry shook their heads in disbelief. It would be far cheaper for companies to replace the door and valve handles and make a few modifications for the already existing, less complicated (and more reliable) robots, than to build and maintain such an expensive, complex, and difficult to control robot. Plus teleoperated cars already existed in 2015 and autonomous cars were clearly on the way, so why would the robot have to drive the car the way a human used to do? They viewed the DARPA Robotics Challenge as a way of trying to create Terminator types of military robots but hiding it under a “search and rescue” glamour. Humanoids are cool and excitingly hard from a research perspective, but not they are necessarily cool and elegant from an engineering design for an end-use perspective.

Q.U.R. was also realistic in considering the product lifecycle costs of a robot and why simple and elegant design is economically superior to complex designs. A major selling point of the Q.U.R. robots to the public was that they would be less expensive to begin with, but would also be cheaper to maintain, because there were less joints, sensors, and effectors to break or jam from lack of use. A major selling point of Q.U.R. robots to the government, which had granted Robinc a monopoly on robots, was that the usuform robots required less resources to build. Expensive metals and electronics would not be put on a robot that would not use those resources— no extra arms or legs consuming valuable materials just because it looked good. Thus, the usuform robots would be what we’d call more “green."

One thing the story Q.U.R. is not is a story about learning from demonstration. I had read Q.U.R. years ago, but recently reread Q.U.R. while researching stories that explicitly detailed machine learning in scifi robots for my Science Robotics science fiction/science fact monthly articles. The technovelgy.com site listed the story as being about learning by demonstration. Learning by demonstration has been a longstanding goal of roboticists; the idea is for the robot to observe a person performing a task and then being able to do it themselves. In Q.U.R., the designers use what would now be called a motion capture system to watch how a tentacled Martian bartender creates a popular drink. Quinby then watches the video and designs a non-anthropomorphic robot to duplicate those motions. If the designers had built a robot first and had the robot watch and duplicate the motion or had a mocap system capture the data and then convert for the robot, then it would have been learning by demonstration. But since it is Quinby learning from the demonstration and then using his analysis to build a robot, it is not machine learning. It is a good idea, just not machine learning.

Don’t allow the madcap bonhomie style of writing distract from the thoughtful insights into robot design, particularly morphology. Get a copy of Q.U.R. from one of the many anthologies it appears in - the Internet Speculative Fiction Database lists at least six different volumes, your library is bound to have one- and think about robot design from a different perspective. You’ll never look at the DARPA Robotics Challenge the same way again.

- Robin