Dr. Florent Perronnin of Naver Labs Europe, who leads future technology research at Naver (NAVER(035420)), said in a video interview held on the 26th at Naver 1784 in Seongnam, Gyeonggi Province, “Robots that are implemented as actual services have a competitive edge over those that merely put on flashy demonstrations.” While humanoid robots that recently demonstrated kung fu or participated in a half-marathon in China have been attracting attention, Naver explains that it aims for service-oriented robotics used in daily life rather than merely showcasing technology.
Dr. Perronnin, a world-renowned scholar in the fields of computer vision and artificial intelligence (AI), joined Naver Labs Europe in 2017 after serving at Panasonic, the Xerox Research Center, and as the head of Facebook AI Research. He has won numerous awards at international AI competitions such as ImageNet and PASCAL VOC, and his 2010 research on large-scale image classification was recognized for its academic impact, earning him the 2020 “Koenderink Test of Time Award” from the European Conference on Computer Vision (ECCV).
Located in Grenoble, France, Naver Labs Europe was established based on the Xerox Research Center, which Naver acquired in 2017. Currently, researchers from 26 countries are studying physical AI and spatial AI for robots, leading the development of next-generation robotics technology.
“The AI competition is now about services, not models”
Dr. Peronin noted that the pace of performance improvements in AI models has recently slowed compared to the past. “Over the past two to three years, model sizes haven’t grown as rapidly as they did before,” he said, adding, “We have now entered the era of ultra-large models utilizing trillions of parameters.”
He cited the lack of high-quality data as the primary reason. “As the saying ‘garbage in, garbage out’ goes, data quality directly determines the quality of the model,” he emphasized, adding, “Securing good data—and ensuring it can be used legally—is a crucial competitive advantage.”
For this reason, Naver Labs Europe adopted a strategy of focusing on robotics rather than covering all AI fields. The company explained that its unique competitive advantage lies in its ability to secure real-world service data accumulated daily by the autonomous robot “Rookie” at Naver 1784, the world’s first robot-friendly building.
Dr. Peronin said, “Focusing on AI robotics was the right choice, given that we can secure real-world robot service environments and data that are difficult for competitors to access.”
When the concept of “foundation models” emerged in 2021, Naver Labs Europe began research on applying them to robots. One of its flagship achievements, the 3D vision foundation model “DUSt3R,” is a technology that reconstructs a space in three dimensions using just one or two photos, helping robots quickly understand new environments. Since its announcement in 2023, it has been cited over 1,600 times, establishing itself as a leading technology in the field of 3D vision, and is being utilized in follow-up research by companies such as Meta, Google DeepMind, and NVIDIA.
He predicted that the core of future AI competition will lie in services rather than the models themselves. Citing Anthropic as an example, he said, “While a good model is important, the company was able to grow rapidly because its user-facing services—such as ‘Claude Code’—were exceptional,” adding, “Ultimately, the winners will be companies that create differentiated services based on their models.”
This aligns with Naver’s “On-Service AI” strategy. Naver is applying its own hyper-large AI, HyperClova X, to search, maps, commerce, cloud, and robotics, thereby establishing a virtuous cycle in which data accumulated from actual services is used to further advance the AI.
“User Value Is More Important Than Flashy Demos”Dr. Peronin repeatedly emphasized that in robotics, too, the key to success lies not in flashy technology demonstrations but in user value.
Citing an instance where a U.S. media outlet described Naver 1784’s autonomous robot “Rookie” as a “trash can on wheels,” he said, “While that’s somewhat insulting, and its appearance may not be flashy, Rookie actually provides useful services to people.”
He went on to explain, “Operating the service allows us to continuously gather real-world user feedback and data, which in turn leads to a virtuous cycle of improving both the AI and the service.”
This is supported by Naver’s multi-robot control system, “ARC (AI-Robot-Cloud).” The core of the system is that the cloud integrates and manages the movement and tasks of multiple robots within a building in real time, reducing the computational burden on the robots and increasing operational efficiency.
He identified real-time performance, safety, and data efficiency as key challenges in robot commercialization. “While a chatbot can afford a one-second delay in its response, even a one-second delay in a robot’s decision-making can lead to an accident,” he said. “Since robotics lacks the vast amounts of training data available for the internet, we need technology that can learn like humans even with limited data.”
The result of these efforts is “DIVINE,” a general-purpose encoder recently unveiled. DIVINE integrates 2D and 3D information into a single model, enabling robots to make real-time decisions even with limited memory and computational resources. Naver Labs Europe plans to expand support to include language and audio by the end of the year and apply the technology to actual robot services.
“Sovereign AI Is Becoming More Important… Korea and Europe Must Move Forward Together”Dr. Peronin assessed that the importance of “sovereign AI” has grown even more in light of the recent controversy over the U.S. government’s restrictions on access to Anthropic’s top-tier AI models. He stated, “Sovereign AI is now a very important task,” adding, “Independence is necessary not only in AI but also across all core technologies, such as GPS.”
He went on to emphasize, “Just as Europe has built its own satellite navigation system, it is becoming increasingly important to reduce dependence on U.S. technology and secure self-reliance in core technologies.”
He cited the French AI startup Mistral AI as a prime example. “The founders are researchers who previously worked together at Meta, and they established the company based on France’s excellent education system and their experience with global big tech,” he explained, adding, “Although Europe faces the limitation of a fragmented market, it possesses a healthy AI ecosystem where outstanding talent, large corporations, startups, and industry-academia collaboration are organically interconnected.”
Regarding South Korea’s “AI Big Three” strategy, he also stressed that cooperation is more important than competing for rankings. “Rather than focusing on the number ‘three,’ it is far more important for nations that share the same values to collaborate,” he said. “If South Korea, Europe, and Canada join forces, we can present a meaningful alternative to the U.S.- and China-centered AI hegemony.”
He explained that these values were also the reason he chose Naver. “Naver develops AI in a way that respects users and has a philosophy of combining each other’s strengths,” he said, adding, “Cooperation between countries and companies that share these values will become even more important in the future.”
which survived by competing with Google… The founder’s philosophy is what moved me”
Dr. Perronin’s decision to continue his research at Naver Labs Europe for nearly a decade was driven by his respect for Naver, which had successfully defended its independent platform while competing against global Big Tech companies.
“France also tried to create its own search service but ultimately failed, and even now, most people use Google,” he said. “Although South Korea is a smaller market than France, it has competed with global Big Tech and preserved its own platform. I wanted to work for a company like that.”
The decisive turning point was his meeting with founder Lee Jae-jin. “The founder told me he created Naver because he wanted to preserve Korean culture,” he recalled.
He continued, “Even back in 1999, there were American portals like Yahoo, but they weren’t services optimized for the Korean language and culture,” adding, “The vision of using technology to help people access their own language and culture more easily struck me as truly beautiful.”
Dr. Peronin said, “I wanted to contribute to that mission, and the founder’s passion remains my greatest source of motivation to this day.”
