Niantic Spatial Develops Spatial AI for Real-World Robots
In 2016, people worldwide took to streets and parks with smartphones to search for Pokémon through the augmented reality game Pokémon Go. The game brought global attention to Niantic, the company behind it. The company is now applying technology developed through its work with real-world spaces to artificial intelligence systems designed for robots.
Cho In-hee, also known as In-hee Cho Su, joined Niantic as chief executive in March. She met in Seoul on 21 September and described the company’s focus on spatial AI. The technology is being developed to create spatial AI models that let humans and machines work safely and efficiently in the same physical spaces. Cho also linked spatial intelligence to the development of true general-purpose artificial intelligence (AGI).
Last year, Niantic sold its gaming business, including Pokémon Go, and restructured as an independent company called Niantic Spatial. The company now focuses on spatial AI, extending the physical-space expertise it developed during the Pokémon Go era into artificial intelligence. Niantic founder John Hanke is a specialist in the field and contributed to Google Earth and Google Maps.
The company’s earlier mapping work was largely designed to help people navigate physical locations. Its current work involves creating 3D and 4D world models that both humans and machines can understand. The approach intends to represent physical environments in a form that artificial intelligence systems and robots can use.
Cho has more than 20 years of experience at IBM and previously served as president of DocuSign, the electronic signature company. Her experience in enterprise software and artificial intelligence led her to observe a limitation in language-based AI. Such systems had focused heavily on language and the digital economy, while physical spaces where people work and machines operate presented a different challenge.
The physical world is also a significant part of the global economy. Cho said about 80% of global GDP is generated in the physical world. She described enabling AI to understand these environments as a harder problem and a larger opportunity than areas that have received more attention from language AI.
Niantic Spatial’s core technology converts real-world locations into three-dimensional models that AI systems can understand. Factories and cities can be captured using 360-degree cameras and reconstructed as 3D spaces. These models can include equipment, pathways and the positions of obstacles within the environment.
The approach lets robots pre-learn real physical environments rather than depend solely on computer-generated virtual spaces. Niantic Spatial’s models therefore begin with locations that already exist in the physical world. This differs from world models that generate places that could exist but are not based on an actual location.
Niantic Spatial can also complete the process relatively quickly. According to the company, a factory can be scanned in 10 to 15 minutes, with the resulting 3D model generated within an hour. The technology is designed to give robots a digital representation of the environment before they operate within the physical location.
Niantic Spatial is also developing technology to help robots establish their position and orientation. Its Visual Positioning System, or VPS, works differently from GPS. GPS uses satellite signals to calculate a position, while VPS compares the view from a robot’s camera with a pre-built 3D map.
This comparison can provide information about both the robot’s location and its orientation. The distinction matters in dense urban environments, where the company says GPS can be inaccurate by 10 to 40 metres. Niantic Spatial says its VPS can achieve centimetre-level accuracy.
The company has begun applying its spatial AI technology through partnerships in robotics and mapping. It recently partnered with humanoid robot company Flexeon to train robots using 3D reconstructions of real offices. The training uses representations of existing physical spaces rather than relying only on computer-generated environments.
Niantic Spatial also signed a memorandum of understanding with Singapore’s Land Authority to build a nationwide VPS based on 3D data. The agreement is part of the company’s wider work on using three-dimensional information to represent physical locations for machine-based applications.
In South Korea, Niantic Spatial is seeking local partners to develop spatial AI and robotics. The company has identified robotics companies, industrial firms and the government as potential collaborators. South Korea has established strengths in manufacturing, engineering, computer science and computer vision, while investment in its robotics ecosystem is growing.
Cho has identified these capabilities as relevant to developing physical AI and said South Korea could play a leading role in the field. The company’s search for local partnerships follows its broader shift from gaming towards spatial AI, with its mapping and visual positioning technologies being applied to physical environments where humans and robots operate.
