Humanoid robots invade Las Vegas at CES, technology touts the future of AI

This is the week Sin City became science fiction.

Humanoid robots shadowboxed, danced, and pretended to run small stores. Singapore-based Sharpa demonstrated a robotic hand playing table tennis and dealing blackjack hands.

Across Las Vegas, technology companies used the annual CES show to reveal their visions for the future and loudly proclaim that physical artificial intelligence is poised for a big year.

“The humanoid industry builds on the work of the AI ​​factories we build for other AI projects,” Nvidia said CEO Jensen Huang during a press conference on Tuesday.

Nvidia, which became the world number one last year most valuable companyannounced a new version of its vision language models called Gr00t for humanoid robots capable of transforming sensor inputs into control of the robot body, as well as a version of its Cosmos model for robot reasoning and planning.

Huang said he expects to see robots with human-level capabilities this year.

“I know how quickly technology moves,” he said. His company has highlighted partnerships with companies like Boston Dynamics, Caterpillar and LG.

Science fiction writers have dreamed of this moment for decades. “The Jetsons” had Rosey, a robot servant. In “Star Wars,” C-3PO helped Luke Skywalker save the galaxy. However, in real life, humanoids have so far been unable to demonstrate the intelligence or flexibility that would make them truly useful, a problem that has long eluded engineers.

Then came generative AI with the launch of OpenAI’s ChatGPT in late 2022. The same deep learning technology that underpins ChatGPT can be used to teach robots to walk, use one hand, or fold laundry. Many in the industry view self-driving cars as the first major commercial manifestation of physical AI.

Industry heavyweights are going big.

In addition to Nvidia, other chipmakers Advanced microdevices And Qualcomm made big announcements about robots at CES. Monday, Google deep mind said it would work with Hyundai’s Boston Dynamics, formerly a division of Google, to develop new AI models for its Atlas robot.

McKinsey estimates that the market for what it calls general-purpose robotics could reach $370 billion by 2040, with key use cases including “warehouse logistics, light industrial, retail operations, agriculture and healthcare.”

But some analysts were quick to point out that there’s a long way from showroom to factory to home.

“Even though it was the humanoids that caught everyone’s attention, and it was the best kind of eye candy of the series, we are still very, very far from their commercial implementation,” Ben Wood, chief analyst at CCS Insight, said in an interview.

AMD CEO Lisa Su on Monday unveiled a new humanoid robot from Italian company Generative Bionics, a company she financially backs. The robot, Gene.01is expected to be deployed later this year in industrial environments like shipyards.

Generative Robotics uses AMD’s cloud-based graphics processing units to train and refine its models.

“This allows us to customize the next generation of models on their GPUs,” said Daniele Pucci, CEO of Generative Robotics. “It’s the brain.”

For now, sales of robotic chips represent only a small fraction of Nvidia’s business, and AMD presents them as “embedded” sales, a term for industrial chips. Qualcomm’s “Internet of Things” revenue accounted for about 18% of the company’s sales last fiscal year.

But they see an opportunity to capture business with a new generation of robot makers by offering them not just chips, but an entire software ecosystem to make development easier.

“This is about one of the major players establishing itself as a one-stop shop for the robotics development community,” Wood said.

As the tech industry has become enamored with large language models in the generative AI boom, many robots are enabled by vision language models. They can combine data from a robot’s sensors with traditional AI models to enable reasoning or planning, such as a route through a floor cluttered with obstacles.

Alongside Nvidia’s VLM announcements at CES, Qualcomm showed off a new line of robotic chips called Dragonwing that can use the company’s VLMs. Qualcomm uses teleoperations to teach its specific VLM skills, such as how to use actuators to grip an object.

A particularly interesting area for Nvidia is medicine.

The company presented a robot from a company called LEM Surgical using its Thor chip. The robot was described as a humanoid, but it had no legs. Instead, it had three arms, two for using tools and one that controlled a face-shaped camera and sensor module. Its sole function is to assist doctors with spinal surgery.

In the exhibit hall, Nvidia showed off a Chinese humanoid robot called Agibot that used a large language model to chat with attendees, although it struggled to stand on the conference center’s plush carpets.

Some of the first humanoid robots on the market may be more about fun and flash than productivity. China Unitree Robotics introduced its $70,000 G1 robot at CES. Large crowds of spectators were treated to a boxing and dancing spectacle on the exhibition hall.

The world’s largest technology companies are betting on rapid market change. Nvidia’s Huang said this week that bots are having their “ChatGPT moment.”

Modar Alaoui, general partner at ALM Ventures, sees robots quickly moving from novelty to reality.

“The next generation is going to grow up with these machines whether we accept it or not,” he said.

— CNBC’s Katie Tarasov contributed to this story.