CES 2026 robots were wild, silly, and sometimes disappointing but there was one that restored my faith in robotics

Every tech category has its moment, and for robots, that moment was CES 2026. There were dozens of humanoid robots (and a fair number of vacuum robots that grew legs) turning the tech event into a sort of carnival of inhuman oddities.

There were boxing robots, card-playing robots, ping-pong-playing robots, dancing robots, robots that did the laundry, bots that sorted parts, and robots wearing elaborate costumes. It was an eye-popping and extraordinary display of engineering and AI-based training.

Image 1 of 4

The unbeatable bots

Even as bots beat me at Tic-Tac-Toe and took my virtual money in a game of blackjack, I enjoyed them all. Still, if I'm being honest, most of them paled in comparison to Boston Dynamics and the first public demonstration of its Atlas humanoid robot.

To understand the stunning achievement that is Atlas, I think we have to categorize and address some of the robot competition.

There were, for instance, quite a few robots with a torso, head, and arms, but lacking legs and instead either fixed or moving about on rolling bases.

I saw two solid examples of this in LG's CLOiD and Switchbot Onero H1, both of which purport to help you sort, do, and fold your laundry.

First, I watched as the wheeled bot CLOiD took the LG CES press conference stage and very slowly grabbed a towel from an LG executive and then carefully placed it in an LG washing machine. It was only impressive because this was the most I'd ever seen CLOiD do.

Next, I watched as Onero H1 attempted to perform a similar task. It grabbed a bright yellow garment from a couch and then rolled over to the washing machine, pulled open the door, and placed the clothing inside. Then the tall, pale robot closed the washing machine door. Unfortunately, the H1 failed to ensure that the clothing was entirely inside the machine, and once the robot tried closing the door, it failed. Worse yet, the robot didn't even notice or attempt to fix its error.

Playing games

At another point, I played chess against Senserobot Chess Mini. It was fun, but do I really need a robot to move chess pieces when I can play on my phone with the Chess app against either the onboard computer or someone else around the world?

Even the relatively skilled Ping Pong robot I saw was still attached to a giant immovable base, and I knew immediately it would be no match for Marty Supreme

Walk it out

More realistically humanoid and bipedal robots, on the other hand, ranged from simple sorting routines (red balls go on red plates and blue balls go on the blue) to gimmickry. I saw far too many robots dancing like awkward middle-aged dads at their daughter's wedding, and a bizarre number of robots that enjoyed performing a good roundhouse kick.

Engine AI showed off some impressive, pint-sized humanoids who could tumble, dance, and queue up to box each other. Weirdly, no robot ever landed a punch. Metal and gear Rocky Balboas, they were not.

Sometimes the robots fell over or failed in their tasks. I played Tic-Tac-Toe against a Fourier robot, which could not seem to place the foam cube playing pieces in the allotted spaces. Sometimes, I impulsively reached in to properly position a piece. Sure, it beat me, but did it do it neatly?

I shook hands with an awkwardly shuffling Zeroth M1 robot, but like so many of them, it appeared to be teleoperated. And then there was Limix Dynamic's Tron, a squat little robot dressed in a full-sized animatronic dinosaur costume. It was kind of amazing but also pointless. Dynamic's Olli humanoid showed off some smooth dance moves but offered no indication that it could do much without someone pulling the digital strings in the background.

The Atlas difference

It's not clear how long any of these companies have been working on these robots, though I think it's clear that some of their training has, in recent months, been accelerated by AI.

Boston Dynamics' Atlas is clearly also benefiting from generative AI and fresh models backed by NVIDIA's platform (and soon Google's Gemini), but it is also an old robot soul compared to these newcomers.

I first saw Atlas in person (or at least a version of it) more than a decade ago at a DARPA robotics competition. Those bots were huge, heavy, awkward (a little scary), and struggled to climb over obstacles and perform basic tasks like opening a door. As one Boston Dynamics exec reminded me last week at CES, "It couldn't do anything."

The all-electric Atlas, especially the new production model on display at CES 2026, is an entirely different beast.

In the demonstration I saw, it was laid out flat on the ground in an awkward position that I would struggle to get up from. Atlas effortlessly stood up (while moving its joints and body in ways no human ever could) and then casually stepped forward. There were waves, dances, and poses that reminded me of bodybuilders in a pose-down. In other demos, it showed off how it could perform factory work.

None of Atlas's skills appeared to be gimmicks or built just specifically for the day. Atlas has a better sense of its surroundings and more command of its skills than the majority of robots on display at CES.

It's also built to head to factories, not homes. Most of the humanoid robots that might end up in homes this year were not at the show. Tesla Optimus showed up briefly but left with Elon Musk. 1X Neo was not around, nor was Figure AI's Figure 03.

Even though they're designed for different markets, Atlas is probably the truest representation of the current state of humanoid robots. It may, in fact, be the best of them. It does not overpromise and underdeliver as I think many of these home robots surely will. Instead, it's headed to a Hyundai manufacturing facility (Hyundai is now a major Boston Dynamics investor) in 2028. There is no timeline for the home, but when it's ready, I bet Atlas is the first home humanoid robot that doesn't disappoint.