China's Unitree CEO sees GPT-like robotics breakthrough soon

Wang Xingxing, founder of China’s Unitree Robotics, said the robotics industry could see a breakthrough similar to ChatGPT within the next two years if AI advances further.

Speaking at the World Robot Conference in Beijing, he described this as the point when robots can perform tasks in unfamiliar places without prior training.

While hardware and training data are sufficient, he said AI for embodied intelligence is still lacking.

Wang expressed doubts about the effectiveness of popular vision language action models but said other approaches, such as using generated video models, might hold more promise.

He pointed to Google’s Genie 3 model as an example of progress.

Founded in 2016, Unitree started developing humanoid robots in 2023 amid rapid AI advances.

The company recently signed a 46 million yuan (US$6.4 million) deal to supply robots to China Mobile and is preparing documents for a potential stock market listing in China.

🔗 Source: South China Morning Post

🧠 Food for thought

1️⃣ Robotics breakthroughs have historically taken decades longer than predicted

Wang’s prediction of a robotics “ChatGPT moment” within two years follows a pattern of ambitious timelines that robotics has struggled to meet throughout its history.

The first industrial robot, Unimate, was developed in the 1950s and used on General Motors’ assembly lines, but it took until the 1980s for robots like PUMA to achieve complex assembly tasks1.

Even then, these systems required extensive programming and couldn’t adapt to unfamiliar environments, which is the challenge Wang identifies as needing breakthrough AI for “embodied intelligence.”

Mobile robots faced similar delays: Shakey, created in the 1960s as the first mobile robot capable of reasoning about its actions, represented a major milestone, but truly autonomous mobile robots only emerged decades later2.

The gap between robotics demonstrations and real-world deployment has consistently been wider than initial projections, suggesting Wang’s two-year timeline may face similar challenges despite current AI advances.