Robots are Inside Out

Every human body contains an infrastructure to build and maintain that body. Robots have a similar structure, but outside the body — not an infrastructure but an
“exostructure.”

We roboticists often take our inspiration from the human body, but not the whole body. We mostly ignore the internal organs. We focus on the skeleton, the skeletal muscles, and perhaps the skin and some of the sensors.

And, why not? Why should we care about the heart, the bowels, and so on? Robots don’t need them.

But let’s dig deeper. Generally, there is much to learn by comparing robots and humans, and pondering the differences. So, let’s phrase the question more carefully, in more general terms. Animals have an internal infrastructure for distributing nutrients, for maintenance and repair, and even for manufacture, taking the body from infancy to maturity. Why don’t robots need such an infrastructure?

The answer is obvious, once you phrase the question carefully. Robots do need a similar structure, and they have it, but it is external rather than internal. There is a significant part of the world economy, including factories, service people, logistics, power companies, and other resources, without which robots could not operate or even exist.

What do we call the robot’s version of the human’s infrastructure? We need a word analogous to infrastructure, but external. I couldn’t find a good word so I made one up: “exostructure.” A mashup of exoskeleton and infrastructure. (If you don’t like exostructure, how about outards versus innards? Or perhaps you can think of a word to contrast with guts or viscera.)

Humans

If, like me, you are the argumentative type, you are thinking, “But humans have an exostructure too.” Yes, that’s true. Young humans are nurtured by their mothers, protected and trained by their families. Modern humans have a vast number of farms, factories, and a distribution network, and more, providing conveniences that we depend on, and which we share with the robots. In addition, we have doctors, nurses, pharmacies … an entire medical establishment to help with maintenance and repair. But I think we can agree that the human body relies primarily on internal mechanisms for manufacturing, maintenance, and repair.

Let’s do an experiment to examine a human’s relation to the exostructure. Let’s force a human to operate autonomously. We’ll take a human away from their exostructure and see how they do, relying almost entirely on their infrastructure. We do have to give them access to land, and some natural resources for food and water, but we could stop there. We could put the humans in the wild, in an environment that might be similar to where humans and their infrastructures evolved. Then we watch and see how long they survive. We cannot actually conduct that experiment, but we can look for historical examples. For example, Marcos Rodríguez Pantoja, alone from the age of seven, survived for 12 years in the mountains of Spain. And if you are a fan of Naked and Afraid, you have watched many survivalists spend three weeks with minimal exostructure.

So, a human can survive alone in the wild for weeks, months, even years. Even though evolution arranged for humans to be supported by their families and their clans, we still retain the ability to survive alone. A human with no external support can obtain nutrients, convert them to energy and other basic needs, distribute them, construct new components and distribute those, use them to replace worn components, and even grow up.

Can we measure the human infrastructure? It is based on incredibly tiny little widgets—cells and even smaller structures. Weighing it, or measuring volume, would make no sense. There is a lot of fancy chemistry going on, but I don’t know how to measure that. I do know one way to measure the mechanical stuff: let’s count motors, or to be precise, let’s count actuated degrees of freedom. There are a lot of them! That would include each flagellum, a propellor that spins to drive a sperm cell forward, and each cilium, a waving hair that helps propel an ovum down the fallopian tubes, and the similar structures in the lungs and elsewhere, and then also the tiny structures inside each cell, to reshape the cell, to move stuff around, and to divide the cell. Each cell has its own infrastructure! If we add it all up, first there are the 200 – 300 skeletal degrees of freedom that we roboticists are so fond of, and then for the infrastructure there are trillions, or maybe quadrillions, of actuated degrees of freedom. To say the visible part of the human body constitutes just the tip of the iceberg would be a gross understatement. The infrastructure is indeed a vast and complex system.

Robots

Now let’s look at robots, and robot exostructure. The robot’s exostructure includes human attendants who step in to correct failures, or make repairs, as well as preventative maintenance. Then, going back in time, the engineers and mechanics who designed and implemented the installation. And going back further, the manufacture of the robot, and its components, and its components’ components.

While each human body has its own infrastructure, the robot’s exostructure is shared across many robots, and also, to some extent, shared with non-robotic systems and even humans. Counting actuated degrees of freedom does not make much sense. Another possible measure would be dollars, but I cannot figure out how to do that.

Here is a relevant measure: Mean Time Between Interventions, abbreviated MTBI. How long, on average, can a robot do its job autonomously, without human interventions for maintenance, repair, or failure recovery? This measure misses the manufacturing and deployment processes, but otherwise it works nicely. How long can a robot “survive” in the wild, without access to its exostructure. Naked and Afraid, for robots.

There are complications. First, there is a tradeoff with the demands of the robot’s task. For our human-in-the-wild experiment, the task is to survive, and perhaps to find one’s way out of the wild. But with robots, there is a wide variation. Suppose the task is to hold down a stack of papers, emulating a paperweight. The robot could sit on the stack and turn itself off. The MTBI would be high, perhaps decades or more, perhaps until an earthquake shakes the robot off the stack of papers. At the other extreme, suppose the task is for a humanoid to play short-order cook at a diner. Given present humanoid capabilities, the robot would need an “assistant” at all times, doing all the actual work. The MTBI would be zero.

There is another limitation in using MTBI as a measure. In some cases, the robot is doing lots of complicated stuff, while a human attendant is doing practically nothing, but doing it constantly. The Mars rovers are sometimes given an updated plan on a daily basis by earthbound humans. There are many other cases where robots might be designed to work with humans. In that case you might want to look at unplanned interventions.

For these reasons and others, MTBI figures are not easy to find. But I know that preventative maintenance for some industrial robot arms occurs weekly. And I found a study of a rover patrolling two different industrial facilities that found MTBIs of 78 hours and 140 hours respectively.

Exostructure versus infrastructure

Should we try to build robots comparable to humans, with an infrastructure rather than an exostructure? Why? Let’s continue to spread all those facilities out in space and amortize the costs over many robots. This is a clear advantage of exostructure over infrastructure.

But there are surely some advantages to an infrastructure. One issue comes to mind. Humanoid developers want to put tactile sensors in the skin. Human skin is constantly being repaired from the inside. To maintain tactile sensing skin with the exostructure, robots will probably have to change their skin regularly. There are efforts to make tactile sensing cheap and compact enough that the skin could be renewed in the same way that humans change gloves.

Perhaps the main lesson is that we cannot, and should not, try to copy the human body indiscriminately. We should choose between biology’s solutions and our own technological solutions, component by component. That is obviously true when building specialized robots for manufacturing or warehouse automation, but it is still true even when aiming for general purpose humanoids. Choosing exostructure over infrastructure might be the most compelling departure from nature’s example. Other choices are less clear. We still have much to learn. Let’s take advantage of biology as an inspiration, but let’s keep an open mind and use what we learn to produce the most effective and most elegant solutions.

Notes and details

For more about Marcos Rodríguez Pantoja, the child who survived in the wild for 15 years, see “How to be human: the man who was raised by wolves,” The Guardian, August 28, 2018. Link

Here is the autonomy study I mentioned: Staniaszek, Michal, et al. “AutoInspect: Towards Long-Term Autonomous Industrial Inspection.” arXiv preprint arXiv:2404.12785, 2024. Link

Regarding planetary rovers, here is a detailed study of autonomy measures: Schreckenghost, Debra et al. “Measuring robot performance in real- time for nasa robotic reconnaissance operations.” In Proceedings of the 9th workshop on performance metrics for intelligent systems, 194–202. 2009. Link

For a survey of tactile sensing skin, see: Roberts, Peter, et al. “Soft Tactile Sensing Skins for Robotics.” Current Robotics Reports 2 (2021): 343–354. Link