We are living in the midst of a profound technological restructuring of human society. The machines that once only frolicked in science fiction have begun to infiltrate our lives. If you don’t already work alongside a robot, you may in the near future. Self-driving cars promise to transform our roads, and the first truly sophisticated robots have begun laboring in hospitals and construction sites and even Walmart.
But behind the autonomous revolution is a mountain of problems. Well, challenges, if you want to be more optimistic. To that end, a panel of roboticists have laid out the 10 biggest challenges for the field in the journal Science Robotics, challenges that touch on a fascinating array of fields. New motors from electrical engineers, new materials from the materials scientists, and even ethical guidelines from the social scientists. Where exactly the robot revolution is headed is unclear, but what’s certain is that it will impact a slew of scientific disciplines.
“We want to use this as the starting point for such a diverse field of research to bring people together to think different,” says lead author Guang-Zhong Yang, a roboticist at Imperial College London.
Body of Work
Let’s start with the physical stuff, the hardware. The panel didn’t worry itself with the challenges of specific kinds of robots, like humanoids or collaborative robots. “This was done intentionally,” says Yang, “because we sometimes pay too much attention to embodiments rather than thinking more fundamentally how we can do differently, how we can learn from nature, how we can use new materials.”
Robots are still for the most part the stilted, unfeeling, kinda stuttering machines, largely because of the limitations of materials. But that’s beginning to change. For one, robots are getting cuddlier. In the aptly named field of soft robotics, engineers are developing squishy machines that, for instance, use the flow of oil to change shape. This could lead to robots that are far safer for humans to work with. First, though, engineers will have to overcome challenges like making sure soft robots can heal themselves if punctured. At the moment, one particular soft robotic hand can heal itself fine, but only when someone applies heat for 40 minutes. Ideally a robot would do this on its own at room temperature.
This is particularly important when taking inspiration from what nature has already proved works, known as biomimicry. If you want to replicate a human hand, for instance, you may want to develop a soft material that’s gentle on real humans, yet can repair itself when damaged. And that’s not even the half of it: The hand is a wildly complex instrument packed with muscles and tendons and tiny bones. How might roboticists replicate that to get robots to manipulate with the skill of humans? Well, copying the thing bit-for-bit probably isn’t the answer. The challenge, then, is getting dexterity that rivals the human hand without all the intricacy.
Another good example of biomimicry is a robot named Cassie, which looks like a disembodied pair of bird legs. What’s interesting here is that Cassie’s creators never said, “Oh, bird legs, let’s replicate that.” They worked out what was mathematically most efficient, and that just so happens to look like bird legs. Still, though, the challenge for robots that replicate nature, especially humanoid machines, is also replicating the unparalleled energy efficiency of biological beings. And if you want to mimic something as small as an ant, good luck getting it to move with traditional motors, known as actuators, which tend to be super bulky.
One way around that problem could be thinking of robots less as one-off actors but parts of a distributed whole. Think tiny robots that work together to, say, construct complex structures. Or agricultural robots that collaborate to harvest. That means figuring out how to engineer machines that may be just a few millimeters long.
Back to School, Back to School, to Prove to Humans That I’m Not a Fool
Of course, building tiny interconnected bots also requires algorithms that can coordinate hundreds, if not thousands of machines—which brings us to the software side of the challenges for robotics. While AI is making great strides in the purely digital space, embodying AI is a whole different story.
For instance, an algorithm can quickly teach itself new skills like recognizing objects by trial and error, known as reinforcement learning. But try to make a robot teach itself something like how to complete a children’s puzzle, and the trial and error could take way longer than the rapid iteration allowed for in a purely virtual world. So going forward, the challenge will be getting robots to manipulate novel objects in the real world.
Robots have gotten a lot better at that, as sensors like lidar have become both more powerful and less expensive. Still, robots regularly fall into fountains and nearly run down dogs on sidewalks. So that bears improving.
And that’s to say nothing about how the robots get along with humans. Perhaps the most fascinating challenges facing robotics have to do with how humans will interact with the machines, a field known as human-robot interaction. It seems straightforward—stay out of each other’s way, help a robot up if it falls over, etc.—but it gets tricky, fast. Last year, for instance, a security robot got into trouble for allegedly harassing homeless people in San Francisco.
On the other end of the spectrum, what kinds of bonds will we form with the machines, both emotionally and physically? What if manufacturers exploit these bonds to, say, convince children to buy ever-more-sophisticated robotic dolls? Can you truly love a robot if it is incapable of loving you back? Ethically speaking, the robotic future looks kinda confusing.
Surely, the majority of human-robot interactions will be more innocent, and indeed they’ll be the norm for many industries. Surgeons already work with robots like the da Vinci system, but in the future the challenge will be handing more and more responsibilities to the machines to do tedious tasks like stitching wounds. That means a truly delicate dance of human-robot interaction, a surgeon working right alongside machines without anyone getting in each other’s way.
Really, who isn’t going to be involved in robotics in the near future? “If you look at our 10 challenges,” says Yang, “you have materials from materials science, the power from electronic engineering, and the navigation control from computer science, hardware systems from biology.” Oh, and ethicists, and neuroscientists for the brain interfaces, and security folks to make sure your new humanoid robot doesn’t get hacked and go on a rampage. It’s shaping up to be a real family affair.