Among people who understand that the robots are coming, there’s a big unanswered question kicking around: what are they going to look like? Will they be friendly faces, or impersonal hunks of metal with mechanical arms and sensors making them look more like forklifts or skid steer loaders?

I saw an excellent panel at The Next Endeavor in Mountain View, CA, this September, (Disclaimer: With Imagination in Action, I help to put on these events), where no less than four business leaders involved in the industry were interviewed by pro reporter Krystal Hu, about how robots are emerging in America today, and what things are due to look like soon.

The Humans and the Robots Should Be Friends

Early in the talk, the group discussed some of the ways that robots will be integrated into human society.

“I think there are multiple layers of robot-human interaction,” said Brad Porter of Collaborative Robotics. “You can start from fenced-off, right.”

He noted the design of Amazon robotic fields, where robotic mops and vacuum cleaners are working in airports alongside humans, doing what he called “turn-taking work.”

“We do this at Mayo Clinic,” he said, giving another practical example, “where the staff hands off a cart, the robot comes and grabs a cart and takes it.”

Then, he explained, there’s joint work, a closer collaboration.

“Maybe you’ve got a robotic surgeon collaborating with a human surgeon,” he added. “I think that progression requires expectation and trust management, right? In order for a human to collaborate, the human needs to understand what the robot’s capabilities are, and what the robot is likely to do. And setting those expectations is not just software and AI, but it’s design.”

This transparency, he said, is paramount.

“We make the wheels visible, so you can see which way the robot’s likely to be turning. If you shroud the wheels, you can’t tell which way it’s going to go. As the AI gets better and better, then the robot can understand better and better what the human is doing, and have a better expectation to model for the human, and that allows for that kind of joint work.”

“One of the biggest things, and most important things, is safety,” said panelist Aaron Edsinger of Hello Robot. “You really have to get that right. You have to design for it from the get-go.”

He talked about strategies at his company, which works on consumer-facing technologies.

“We want to get invited in the door, but we also want to get asked to stay,” he said.

In order to achieve that, he suggested, the vendor has to really consider the realities around the use of a humanoid AI robot in the home.

“It’s very challenging to make it safe, fundamentally, and we should be aware of that, and then be thoughtful about how we’re going about it,” he said.

There was also some discussion of practical design.

One common finding was that the companies involved do not choose to build their robots as classic humanoids.

Former Sony AI president Michael Spranger talked about designing an autonomous table tennis player with a “high frequency control” and a decoy trajectory for safety.

Edsinger described Hello Robot’s design as “a large Roomba with an arm.”

“It’s got a long, skinny mast, very slender, very minimal,” he said. “We call it robotic cubism. Rather than trying to create a human form, we decomposed the human morphology, thought carefully about what it is that we need the form to do, and reassembled it into something. If you squint at it, maybe there’s a humanoid in there, but (we’re) being very, very pragmatic.”

Edsinger noted that non-humanoid forms are generally lower cost, and easier to engineer.

“We’re quite deliberate in not doing humanoids either,” Porter said, explaining that a team at Amazon studied it, but did find that design to be optimal for any of several dozen tasks.

“At no place at Amazon did legs confer any advantage,” he said. “They were only a liability. They were power-hungry.”

To make a robot safer, he said, use a low center of mass.

Here’s another very interesting design element that Porter went into in detail. Essentially, he explained, the company puts the robot eyes lower toward the ground, specifically to signal that the robot is of less than human capability.

“We keep the eyes lower because we’re not signaling that we have some human level intelligence,” he said. “We signal more that we have, say, seeing-eye-dog-level intelligence; that’s communicating an expectation to a human as to how capable this robot should be. When we really want to signal that this thing can speak 36 languages, and do anything that you want, and probably is as expert in cancer diagnosis as the doctor in the hospital, then maybe the eyes come way up, right?”

Panelist Alison Sander at AltWheels talked about a real robust proliferation of robot forms, where the humanoid definitely does not dominate.

“To me, this is one of the fascinating mysteries of the robotic space,” she said. “Everybody’s looking for the humanoid robot, but really there’s been a kind of Cambrian explosion of new robotic forms. So there are robotic honeybees, there are robotic fish, Waymo uses robotics to get us around. So everyone’s waiting for the robots to arrive, but they’ve already arrived.”

Another build that Spranger talked about struck me as, well, very different. It’s also profoundly relevant to discussions around AI swarms and how they may be used in future years.

“We want robots to operate potentially in your body and the veins,” he said, “so they’re going to be very, very small. You know, they’re certainly not going to be humanoids.”

“I think we haven’t really seen people apply the level of intelligence that multimodal models have into an embodied form yet,” Porter said, in response to a question about advancements to date. He mentioned a nursing project, a fall detector, that pushes capabilities closer to that edge.

Edsinger commented on the advent of open weight models and robotics.

“In the robot foundation model space,” he said, “we believe … (it) is going to get commoditized at a level that these open weight models, while maybe they’re a little bit older, a little bit behind, they will become quite, quite good, and so then the value is in in integrating these into a product that people get a clear ROI from.”

“I’m quite impressed by the combination of simulation and reinforcement learning and especially in dynamic tasks,” Spranger added. “All of these backflip robots that you see out of China, they’re essentially trained in simulation, and then put on a real robot, and they’re solving a pretty complex dynamic task.”

Sander brought up the geopolitical connection.

“You don’t really hear any metric for: ‘where are robotics, U.S. versus China’,” she said, “and it’s probably significant.”

Porter noted some challenges from the American side.

“From a pure kind of unit economic standpoint, it is hard to compete with China, from a manufacturing cost of goods standpoint,” he said, acknowledging that there are still opportunities to innovate downstream. “It’s not yet a race to the lowest cost, but you do have to hit the unit economics, to make sure your solution works, and I think you can do that and build in the U.S.”

“It’s very hard to source components that are made in the U.S. that can work for us,” Edsinger said, noting that some components are not made in the U.S. at all.

In closing, Hu asked each panelist for predictions in a sort of lightning round.

“We’re going to see more and more robots,” Porter said, reminiscing about seeing a robot clean at Heathrow airport, “And I think that they’re going to blend in a fair bit. … I think our world will transform that way, in a very quiet but quick way.”

“Robots coming into the home will be (received) through the people that need it the most, which will be, first, people with mobility impairments, and older adults,” Edsinger said.

Spranger had some thoughts on workforce automation.

“In ten years, half to (a) larger percentage of the workforce is going to be gone,” he said, “and so I think there’s such a huge pressure demographically, also financially and economically, that these robots are going to be deployed in factories and really automate more of the things that are not automated today. …. I really think people are underestimating the value of IP and character in robots. And I’d like to see, you know, a lot more emphasis on how robots look like, what their movements are, and not just the utility, but sort of also how they’re contributing to the joy and fun in our lives.”

“We’ve been tracking uh HITL,” added Sander, “so from humans in the loop, to humans on the loop, to humans out of the loop. So, there’s sort of a shift from HITL to HOT. And that’s going to make a huge difference. So all the Waymo cars that are all around us, they’ve gone 220 million miles. That’s humans out of the loop. And whether that happens in drone spotting and drone deliveries, that’s going to be important.”

So, more robots, more different kinds of robots, and drones, and humans increasingly taking a back seat. It’s going to be a lot to get used to. But there we are. Stay tuned for more out of this exciting summit.