Plus One Robotics CEO Discusses Human Oversight in Robotics

The rapid development of humanoid robots has intensified debate over how autonomous robots should ultimately become – and whether removing people from the loop should really be the goal.

Plus One Robotics takes a different view. The company, which develops AI-powered robotic automation for logistics and parcel-handling operations, has long advocated a human-in-the-loop model in which robots handle repetitive work autonomously while trained people remain available to resolve exceptions and make judgment calls when machines encounter situations they cannot handle.

That philosophy is summarized by a phrase frequently used by Plus One CEO Erik Nieves: “Robots work, People rule.”

Recent research commissioned by the company suggests that the public may be receptive to that approach. Plus One’s survey found that 26 percent of Americans see a lack of human judgment as their biggest concern about humanoid robots, while 57.5 percent of respondents uncomfortable with particular robotic tasks say they would feel more comfortable if a trained person could monitor the robot and intervene when necessary.

In this interview with Robotics & Automation News, Nieves argues that advances in humanoid robotics and AI have strengthened rather than weakened the case for human oversight.

He also draws an important distinction between the increasingly impressive intelligence of general-purpose robots and the physical realities of industrial automation.

Foundation models may dramatically reduce the time required to deploy or redeploy robots, he says, but they do not necessarily make machines physically faster.

For high-volume applications such as parcel induction and depalletizing, Nieves expects purpose-built industrial automation to retain a significant economic and performance advantage.

Humanoids, meanwhile, could establish a genuine role in high-mix, lower-volume work where tasks change frequently and dedicated automation is difficult to justify.

Looking five years ahead, Nieves expects supervised autonomy to become increasingly important, with individual human supervisors overseeing larger fleets as robots learn to resolve more exceptions themselves.

But his central argument remains unchanged: the real test of an autonomous robot is not how it performs when everything goes according to plan, but whether it can be trusted on its “worst day”.

Interview with Erik Nieves

Robotics & Automation News: Has the rapid development of humanoid robots changed your view on the need for human oversight?

Erik Nieves: If anything, the rapid development of humanoid robots reinforces my belief that human oversight is still necessary even with autonomous systems. Humanoids are getting better hardware and improved AI, and that’s exciting.

But every new capability brings robots into new environments which are bound to contain exceptions nobody designed for. The best way to test a robot is to see what happens on its worst day, when something shows up that it hasn’t seen before.

At Plus One, we say “Robots work, People rule”. The robot handles the routine, and a person handles the judgment calls. I expect the ratio of robots to supervisors to improve quite a bit, but keeping humans in the loop will only become more valuable.

R&AN: Is the need for human supervision a limitation of today’s technology, or will it remain a permanent part of advanced robotics?

EN: It’s some of both. Today’s AI still struggles to identify its limitations, so having a human backstop is mandatory. Over time and as they ingest more data, robots will ask for help less often, and fewer people will be able to oversee more machines. But the trust part is permanent.

In our research, 26 percent of Americans said their biggest concern about humanoids is a lack of human judgment. And of the people who are uncomfortable with certain robot tasks, 58 percent said they’d feel better if a trained human could monitor and intervene if needed. People want to know that someone is accountable.

Autopilot handles most of modern flight, but we still put pilots in the cockpit. I expect supervision to work the same way: rarely needed, but always ready and available to step in.

R&AN: Are general-purpose humanoid robots technically and economically realistic compared with purpose-built automation?

EN: On the software side, the progress is faster than I expected. Over a decade ago, my team at Yaskawa built a mobile system with two arms, configurable hands and cameras. Every sequence and waypoint had to be taught by hand, and it took weeks to program.

Using today’s tools, that timeline would have been cut down to hours or a few days. Here’s the thing, though: none of these improvements in programming would have made that robot run any faster or made more parts per day because the process flow and the physics of the equipment determine that.

So the promise of robotics foundation models is faster deployment, but not necessarily faster robots. Faster deployment, or really, redeployment, makes a huge difference in operations with higher mix and lower volume.

But in a parcel hub handling higher volumes where the focus is on the robots hitting the rate and uptime they need, purpose-built automation still wins.

R&AN: Where do humanoids offer a genuine advantage over purpose-built automation in warehouses and logistics?

EN: Humanoids still have the advantage when the task changes often, there’s lower volume and the facility can’t be rebuilt around a machine. For example, tending several different stations over the course of a single shift, or work that is moderately repetitive, but not enough to justify a dedicated cell.

When you’re talking about applications that run around the clock, like parcel induction or depalletizing, a fixed arm with good vision and human oversight will be faster, cheaper and easier to maintain. I know you reported on our eight-hour livestream where we ran at nearly 2,500 picks per hour. That’s the bar a general-purpose machine has to clear.

R&AN: Is the fact that workplaces are designed around humans a compelling argument for humanoid robots, or should automation be designed around the task?

EN: The argument that the world is built for human bodies is reasonable, but it’s also overstated. Most warehouses are designed around forklifts, conveyors and pallets, with people working around them. A lot of what makes a building “human compatible” is flat floors and wide doorways, which wheels handle just fine.

I’m encouraged by the new wave of wheeled humanoids from Dexmate, Cobot and Apptronik. We’re watching Gil Pratt’s robotic Cambrian Explosion, with robot forms evolving to fit the environment and the job.

My advice is to start by evaluating the job that needs to be done. If legs or a human form help, great. If they add cost, complexity and failure points the task doesn’t require, design the automation around the task instead.

R&AN: Is the robotics industry pushing autonomy faster than either the technology or the public is ready for?

EN: Yes, our survey found that people are uncomfortable with robots in some areas: 52 percent said watching children, 46 percent said performing surgery and 39 percent said caring for elderly people. In those situations, a mistake can mean a person getting hurt, which is entirely different from a package getting sorted incorrectly.

Surgical robotics has handled this by keeping a surgeon in control. My concern is home and care robots marketed as autonomous before they can reliably handle unpredictable human environments. Warehouses are a great proving ground in the meantime because the stakes are lower and the variability is more contained.

The industry can earn trust by proving reliability there first, keeping humans in control in high-stakes settings and being honest about what the machine can and can’t do yet.

R&AN: In five years, what role do you expect humanoids, supervised robots and conventional industrial automation to play?

EN: In five years, I predict that conventional industrial automation will still do most of the work. As I said earlier, in higher volume settings with defined tasks, the economics of purpose-built systems are tough to beat.

I expect the biggest shift to be supervised autonomy, where we see robots resolving more exceptions on their own and each supervisor overseeing more machines.

There may be some cases where there are fully autonomous operations without humans present, but those will be in very specific, controlled environments. That will remain the case in manufacturing, but I don’t expect them to be the norm in warehouse work.

Humanoids with legs or wheels will have a strong foothold in high-mix industrial work, like machine tending, kitting, or moving material between stations. It all comes down to robots that customers can deploy quickly and trust on their worst day.