* It has been translated using AI

GOH Minseo

esms46@mk.co.kr

Input :

2026-09-27 12:00:00

?? ????

Cho In-hee, CEO of Niantic Spaces, switched to robot AI after selling the game
25 years of space technology created by Google Earth
"Building a 3D model in just one day with a phone
"Building an error centimeter-class robot, manufacturing, and looking for a collaboration partner."

?? ??

Cho In-hee, CEO of Niantic Space, who visited Korea, is giving a interview with Maeil Business Newspaper. <Photo = Niantic Space>

In the summer of 2016, the world poured into the streets. It was an augmented reality (AR) game "Pokemon Go" craze that caught virtual monsters with smartphones. Ten years after its launch, it is still a steady seller (more than 1 billion cumulative downloads) enjoyed in more than 150 countries. However, Niantic, the U.S., which created the game, sold its popular game division to Scopley for $3.5 billion (about 5 trillion won) last year. In addition, it has started anew as an independent corporation called 'Niantic Space' by separating only geospatial artificial intelligence (AI) technology. Cho In-hee, CEO of the company, visited Korea last week in search of a collaborative partner. It is the first time for the Niantic Space to visit Korea.

Advertisement

CEO Cho, who led the AI and big data business at IBM for more than 20 years and served as president of Documentary, an electronic signature company, took office in March this year. It is clear why he, who has been involved in digital software, has turned to the 'physical world'. He said, "Only 20% of the world's economy is dealt with by the first-generation AI that has learned Internet languages like Chatbot, and AI does not yet understand the physical world where the remaining 80% takes place. The next generation of language is multimodal and physical AI." CEO Cho said, "We don't play games, we don't do consumer business anymore. We focus thoroughly on corporate (B2B) customers," he stressed.

?? ??

Niantic Space Company Overview. <Gemina Gemini Image>

AI to read space, just like LLM to read

What Niantic Space makes is the 'Large Geospatial Model (LGM). Just as the Large Language Model (LLM) understands text, it is a platform that allows robots and AI to understand the real 3D space geometrically and meaningfully. If the camera is an "eye," it is more like planting a "space sense" that determines where and what the space that came into your eyes is. CEO Cho said, "Universal artificial intelligence (AGI) does not come without spatial intelligence."

The core technology is summarized in three categories. It is a "Reconstruct" technology that creates a 3D digital twin that is the same as the real thing with photographing data such as cameras and drones, "Localize (VPS)", which positions itself with visual information even in indoor and urban areas where GPS is cut, and "Understand" technology that reads the meaning of objects in space. In addition, a technology that compresses the "Gaussian Splat" (a technique of expressing 3D scenes in the form of countless dots) established by academia so that machines can read it was added. Ultimately, the goal is to build a '4D world model' that even includes changes in time.

The strength is speed and cost. The company makes a 3D space model that used to cost months and tens of thousands of dollars with expensive equipment in a few minutes with a regular 360-degree camera that costs less than $1,000 and makes it in a day. Unlike GPS errors that range up to several tens of meters in the city center, VPS narrows them down to centimeters (cm).

The way data is secured is also different. Unlike competing camps that focus on robot viewpoint (ego centric) data, it combines various spatial data covering the ground and aviation (drone and satellite) up to the seabed. Public data released by the government and private data from companies are used together. CEO Cho said, "This spatial AI gives humans superhuman abilities and expands equally to robots working in the same space."

?? ??

Niantic Space major techniques. <Gemina Gemini Image>

From delivery robots to refineries and humanoids

Technology is already used in the field. Coco Robotics, an American delivery robot company that signed an infrastructure partnership in March this year, supplies VPS to prevent robots from getting lost even in alleys where GPS is not caught. The case of ExxonMobil, an oil refinery unveiled in July, is an industrial site. Image-based positioning replaces the search for valves and facilities where employees were wandering with paper drawings at a factory that was blocked by steel and did not pass GPS.

Advertisement

Humanoids help from the learning stage. In collaboration with Nvidia and Flexion, Niantic changes the real space captured by 360-degree cameras into a virtual environment for robot training (real two-seam), and robots learned only in simulation walk around the office without relearning. Recently, the virtual space has also added an editing function that changes lighting and weather. This is to train robots more widely by creating conditions such as night or rainy days.

It is also expanding its presence in the government and public sectors. Recently, a memorandum of understanding (MOU) was signed with the Singapore Land Authority to establish a positioning system for robots and emergency rescue throughout the city. The scope of application ranges from manufacturing, logistics, and energy to urban planning, defense, and the public.

The technology is already accessible to general users. Using the company's 3D scanning app "Scaniverse," you can 3D view the surrounding space with only a smartphone or 360-degree camera.

Accuracy is what the company puts first. CEO Cho emphasized, "If language AI is wrong, one sentence is only wrong, but if the robot perceives space incorrectly, it leads to an accident."

CEO Cho had high expectations for the Korean market. He cited Korea as the 'world's top five' that will actually deploy advanced robots and AI along with the United States, China, Japan, and Germany. "Korea, a manufacturing powerhouse, has unique hardware and software capabilities, research talents, and high-quality data dealing with complex physical spaces," CEO Cho said. "In particular, the research manpower in the advanced computer vision and multimodal fields is the best in the world compared to the population."

There is no branch office in Korea yet. This visit to Korea is also aimed at finding a collaborative partner. CEO Cho said, "I am confident that Korea will become one of the leaders in physical AI," adding, "I want to share that journey with Korean companies."