Legs not wheels: Apptronik's RL lead makes the safety case for the human form

Legs not wheels: Apptronik's RL lead makes the safety case for the human form

John KoetsierNovember 21, 2026
ShareXLinkedInFacebook
Apptronik reinforcement learning lead Kyle Morgenstein argues the real case for the human form is physics, not door frames: wheeled bases carry huge reflected inertia while legs let robots stay compliant with small actuators. His PhD thesis on humanoid safety has been seven years in the making.

Legs: better than wheels?

The standard pitch for humanoid robots is that they fit through your door and pick up your tools. Kyle Morgenstein, the reinforcement learning lead at Apptronik, says that version of the argument is "the weakest possible version" of it.

The case for legs, he wrote on X on Wednesday, is about physics: specifically what happens when a robot bumps into you.

The kitchen problem

Morgenstein was responding to roboticist Barrett Ames, who had posted that companies claiming the human form suits existing infrastructure "are full of it," and that people want humanoids because the form "grabs their attention."

Morgenstein agreed the infrastructure argument is weak. If you take it literally, he wrote, it means "I built my house for me so the robot shaped like me physically fits."

"Industrial robots could easily do your dishes, but you wouldn't be allowed in your kitchen at the same time as the robot," Morgenstein adds.

In other words, the question is not whether a different kind of robot can open the cabinet. The question is whether a human can be in the room safely.

Wheeled mobile bases, he says, solve the size problem but make the cohabitation problem worse.

"They are actually less safe," he wrote, "because the reflected inertia of the base is huge."

Reflected inertia is the hidden cost

Reflected inertia is the effective mass a person experiences when a robot hits them, including the inertia of the drive motors and gearing.

A wheeled base built to carry a robot of any substantial size brings all of that mass into a collision, plus the inertia of whatever is spinning inside the drivetrain. 1X takes the same reasoning a step further inside the robot joints: at Runtime 2026, CEO Bernt Bornich said NEO avoids harmonic drives and planetary gears in favour of high-torque motors pulling on tendons, so there is not "that much kinetic energy in motion when doing things, because there's no gears that are spinning very fast."

Legs solve it differently. "

Legs give you dynamic stability to catch yourself," Morgenstein wrote, "and you can control your CoM with exponential acceleration even with small actuators." Getting the same reactive compliance on wheels "would require giant dangerous motors and batteries."

The Humanoid Technology Expo is where the builders, integrators, and enterprise buyers shaping this space come together. Early registration is open.


Where wheels do win

Morgenstein did not claim legs win in every situation.

Wheeled bases "are safer in catastrophic failure because they're statically stable and can 'just stop' or e-brake," where a humanoid would fall. His point is about frequency: "in the much more frequent occurrence of bumping into people when working in close proximity, legs are safer."

Several companies are building in the other direction.

Swiss startup Embodied AI pitches its wheeled AIR robot as able to work beside factory staff without a cage. Feather sells a $30,000 wheeled humanoid to developers. Agility Robotics has a wheeled base concept alongside Digit 5.

Sunday Robotics makes the most direct counter-claim.

The company says on its website that its wheeled home robot Memo is "compliant and passively safe" ... "if you bump into it, it naturally yields, and it remains stable even if power is cut at any configuration."

That pairs the collision tolerance Morgenstein attributes to legs with the static stability he concedes to wheels. If that claim holds up at scale, it closes a lot of his argument.

Seven years in the making

Morgenstein said humanoid safety "has been wildly under studied" and that the question is the subject of his PhD thesis, 7 years in progress.

He is a PhD candidate in the Human Centered Robotics Lab at the University of Texas at Austin, advised by Luis Sentis. Apptronik was spun out of the same lab. Before that, he studied aerospace engineering at MIT and did research at NASA's Jet Propulsion Laboratory, Boeing and MIT Lincoln Laboratory, and worked on Ultra-Mobility Vehicles reinforcement learning at the Boston Dynamics AI Institute.

"Much more to come soon," he promised.

It's interesting timing for this debate.

GXO, the logistics operator, signed a letter of intent last week for up to 1,000 Apollo humanoids from Apptronik, with expanded warehouse trials starting in 2027. Formal standards are beginning to form: LG Electronics recently agreed to put its CLOiD robot through a pilot safety certification with Korea's robot certification agency. As humanoids move from demonstrations to warehouses and home kitchens, the physics of contact are going to matter more than the physics of doors.

Sources

Share this article

ShareXLinkedInFacebook

In this story

Follow to get an alert when they're in the news

Apptronik
Company · Humanoid maker
USA
NEO
Robot · 1X Technologies
Commercial · 168 cm · 30 kg
AIR
Robot · Embodied AI
Pilot
Feather
Robot · Feather Robotics
Commercial · 180 cm · 105 kg
Agility Robotics
Company · Humanoid maker
USA
Digit 5
Robot · Agility Robotics
Commercial
Sunday Robotics
Company · Humanoid maker
United States
NASA
Company · Humanoid maker
USA
Boston Dynamics
Company · Humanoid maker
USA
Apollo
Robot · Apptronik
Pilot · 173 cm · 73 kg
LG Electronics
Company · Humanoid maker
South Korea
CLOiD
Robot · LG Electronics
Pilot · 125 cm · 65 kg
Newsletter

Humanoid (almost) Daily

The humanoid robot news that matters, with context from HumanoidMegaHub's tracking. Free, almost every day.

Subscribe on LinkedIn

Keep reading

Written by

John Koetsier
John Koetsier

I tell stories with data and build startups. I've raised angel and VC funds, led software companies and startups, and chronicled the rise of the mobile economy. Currently I write for Forbes, host several podcasts, and am starting up HumanoidMegaHub. I was TUNE's mobile economist and Singular's VP of Insights, built the research team at VentureBeat, and managed teams creating software for partners like Intel and Disney. In past lives I built websites, e-learning platforms, and mobile apps, and, a long time ago, was named to Folio's top 100 of the media industry's "most innovative entrepreneurs and market shaker-uppers.”