I hadn't seen Emiglio in years -- probably not since the mid-'90s when I would catch glimpses of him on TV and sigh wistfully. But then one night a couple months ago, in a bar in Porto, suddenly there he was, in the flesh -- or should I say, the plastic.

Emiglio, you see, is a robot. A knee-high butler with a bulbous white head, cartoon grin and red glowing eyes. I spent my childhood longing for him to bring me fun drinks on his little tray while I reclined on the sofa, glued to back-to-back Animaniacs episodes. Looking back, my desire for Emiglio may well have been the reason I committed myself to a career in writing about technology.

Read more: CNET Is Choosing the Best of CES 2026 Awards

Like many fantasies humans harbor about robots, this one was ill-thought-out. Who would have made the fun little drinks and balanced them on Emiglio's tray? How would he have known where to take them? And who, ultimately, would have helped him navigate two rooms from the kitchen to me?

Many of the issues that kept Emiglio from being a genuinely useful butler are the same limitations real-world robots still struggle with today. As much as I hate to admit it, Emiglio was little more than a glorified remote-control car with a face, needing human assistance to do almost anything.

The same is true of Neo, the humanoid home helper robot that went viral in late October but still requires teleoperation by a human. The two robots are separated by more than 30 years, yet their real-world usefulness and ability to operate autonomously feel similarly disappointing.

A question I ask myself each year as I return to CES -- the giant tech trade show the CNET team decamps to every January -- is when the many robots I've met there will finally prove worthy of a place in our homes.

Overcoming the AI bottleneck (and why VLAs matter)

"The main obstacle between us and genuinely useful home robots is the AI," renowned computer scientist Ben Goertzel said when I sat down with him at the tech-focused Web Summit in Lisbon last month. The physical capabilities of robots have progressed dramatically between Emiglio and Neo. What's holding back Neo, and other home robots, is ultimately the smarts.

The AI breakthroughs we've witnessed over the past few years do pave the way for change in this regard. Large language models developed by companies like OpenAI, Google and Anthropic enable us to have more nuanced conversations with our technology, which is particularly compelling in the case of emotional or companion robots.

It's possible that this year at CES, we could see a company integrate more advanced AI into a robot concept it has already showcased, said Ben Wood, chief analyst at CCS Insight. He suggested Samsung could build on its rolling projector robot, Ballie, by working closely with Google -- as it already does on phones -- to create a next-generation version equipped with Gemini, for example.

"Generative AI is allowing more natural language interaction with intelligent devices," said Wood, but this is the same whether you're talking to a smart speaker, a robot vacuum cleaner or a humanoid.

What's more useful to robots than LLMs are developments in vision-language-action models, which, as the name suggests, are a type of AI that allows for the input of a combination of images and words, and the output of actions. For robots navigating a physical space, this combination is essential and will really set them apart from other AI-equipped devices.

"More advanced models for robotics blended with more capable and more deeply integrated generative AI could see some more intelligent use cases coming to either high-end humanoid science fiction robots or some of the more practical robots," Wood said.

There's been debate over whether a breakthrough for robots will require us to unlock AGI -- a hypothetical level of AI superintelligence. Goertzel, who works both on superintelligence and robotics, doesn't think this is the case. VLAs are getting so good, we don't need AGI to make decent home robots, he told me.

Looking beyond the humanoid form

The big challenge for home robots is that every home is different. 

Places such as hotels, schools and hospitals share enough similarities that robot navigation can be more or less standardized. But developing robots for enterprise and industry, where they'll be doing repetitive tasks in predictable environments, is totally different from training a robot that you could put into the wildly varied layouts of homes.

Still, some are trying. The team at Sunday Robotics, based in California, is training its humanoid robot, Memo, with data provided by families across the US that use high-tech gloves to capture the intricate movement of their hands as they carry out household tasks. It's an ambitious approach to preparing robots for family life, and if Sunday Robotics can stick to its desired timeline, it could well be one of the first companies to deploy non-teleoperated humanoids in people's homes.

But for some, there's a real question mark over whether we should be aiming for humanoid home helpers at all. 

"If I think about just everyday stuff, like housework, a humanoid body is not optimal," Goertzel said. "If I think about in our kitchen at home, my wife needs me to reach the high things, and I don't like crawling around in the ground to get the low things right, because we're slightly different heights. Why build that problem into a robot?"

Instead of equipping one expensive home humanoid with the same impediments we struggle with as humans, he envisions a networked system of smaller, more practical bots that can interact and are designed to excel at specific tasks.

There are opportunities for established tech companies to jump in here, whether that's Samsung with Ballie and Apple with its mysterious rumored home robot plans, or companies such as Qualcomm, which will be at CES and may discuss its own robotics plans at the show. 

Qualcomm already makes chips for cars (which are close relatives of robots, especially in their autonomous form) and a whole range of small consumer electronics that maximize power while providing long battery life and AI capabilities. At Web Summit, CEO Cristiano Amon told me that he saw robotics as "an incredible opportunity." 

"We're excited," he said. "Whether from enterprise to consumer, I think the type of silicon that we develop for phones and for [edge computing] is the perfect silicon for robots."

The practical robot: Task-specific bots

Many of us have already begun to invest in smaller, task-specific bots by buying robot vacuums, mops and lawnmowers -- an established category that'll only boom further off the back of CES 2026. 

"There will be an absolute avalanche of robotic vacuum cleaners and mops and robotic lawn mowers," Wood said. Even so, he noted, you need to have "the right kind of house" for them to work optimally. 

Still, CCS Insight research suggests that 15% of households across the US, UK, Spain, France and Germany intend to buy a robot vacuum cleaner in 2026. They're not the coolest or cutest robots, but they do win out when it comes to their usefulness and what people are actually willing to spend their money on.

As for a humanoid home robot? "Quite frankly, it's years away," Wood said. "People love the idea of it, but it's a long, long, long way from being something that people would have in their homes or even want."

The urgent question of privacy and safety

His prediction aligns with that of Boston Dynamics CEO Robert Playter, who told Euronews at Web Summit that he believes robots won't be in our homes for at least five to 10 years. (This from someone whose company makes the acrobatically inclined Atlas humanoid and the intimidating Big Dog, both of which had the US military sniffing around.)

There are a lot of reasons you might not actually want that futuristic home robot, from practical considerations such as space and usefulness, to bigger concerns around privacy, safety and cost (Neo is priced at $20,000, and Sunday Robotics told me Memo will be a "high-end" product).

In November, robotics researchers at Carnegie Mellon University published a paper declaring that popular AI models are not yet ready to power robots due to issues ranging from bias and discrimination to unsafe physical behavior. 

The study, which analyzed ChatGPT, Gemini, Copilot and HuggingChat, found that most models were willing to approve commands that would compromise people's mobility aids, allow a robot to brandish a knife, take nonconsensual photos or steal credit card information.

"If an AI system is to direct a robot that interacts with vulnerable people, it must be held to standards at least as high as those for a new medical device or pharmaceutical drug," said co-author Rumaisa Azeem, a research assistant in the Civic and Responsible AI Lab at King's College London. There's an urgent need, she said, for comprehensive and routine risk assessments before these AI models are put into robots.

Just a week after the study was published, startup Figure AI was sued by a whistleblower, who warned that the company's humanoid robots could "fracture a human skull."

As compelling as they are to observe, robots such as Elon Musk's Tesla Optimus and Boston Dynamics' stable of bots can be a little unnerving. But if those sorts of robots are on display at CES this week, the focus will likely be on the skills they can bring to industrial settings rather than to the home.

Robotics will be a "mega trend" at the show, according to Wood, but it remains to be seen whether any of the robots on display will rise above robot vacuums, pool cleaners or cute but one-note companion devices to become a true household essential.

As for me, I still dream of owning Emiglio -- even more so after finally meeting him. It's tempting to head to eBay to see if anyone is selling this '90s robot toy that my career is built upon, but maybe it's better to hold out for the real thing. For now, I'll keep hoping that the truly capable (and safe) robot butler of my childhood dreams can one day become a reality.