NAVER D2SF Backs Khameleon to Accelerate Physical AI and Spatial Intelligence

The announcement of NAVER D2SF’s strategic investment in Khameleon marks a significant milestone in the evolution of South Korea’s artificial intelligence landscape. By backing a startup focused specifically on Physical AI, NAVER is signaling a shift in priority from purely digital generative models to embodied intelligence—AI that can perceive, reason, and act within the three-dimensional physical world. This move comes at a time when the global tech industry is pivoting toward World Models and spatial intelligence, recognizing that the next great leap in AI utility will occur not on a screen, but in the physical environments of factories, logistics centers, and urban spaces.

NAVER D2SF (D2 Startup Factory) has long served as the primary vehicle for NAVER’s early-stage technology scouting, and its entry into the Physical AI space with Khameleon is a calculated bet on the future of robotics. Unlike traditional robotics, which relies on rigid, pre-programmed instructions, Physical AI leverages large-scale multimodal models to interpret complex sensory data and execute fluid, adaptive movements. For NAVER, this investment is not merely a financial play but a strategic alignment with its broader ARC (AI-Robot-Cloud) ecosystem. This system, which powers NAVER’s 1784 headquarters—the world’s first robot-friendly building—requires a constant influx of cutting-edge algorithms to manage a fleet of autonomous service robots.

“

The announcement of NAVER D2SF’s strategic investment in Khameleon marks a significant milestone in the evolution of South Korea’s artificial intelligence landscape.

The competitive context of this investment cannot be overstated. Globally, the race for Physical AI is intensifying, with Silicon Valley heavyweights like Physical Intelligence (Pi), Skild AI, and Figure AI attracting billions in capital. These firms are racing to build a general-purpose brain for robots. By securing a stake in Khameleon, NAVER is ensuring that it maintains a competitive edge in the Asian market, shielding its robotics initiatives from being entirely dependent on Western-developed foundation models. Khameleon’s specific technical focus—likely involving vision-language-action models (VLAMs)—could provide the critical software layer needed to make NAVER’s hardware more versatile and commercially viable.

From a technical perspective, the challenge Khameleon is expected to address involves the sim-to-real gap—the difficulty of transferring AI trained in digital simulations to the messy, unpredictable reality of the physical world. Physical AI startups typically focus on creating robust data pipelines that capture real-world interactions, allowing models to learn physics, spatial relationships, and tactile feedback. If Khameleon can successfully refine these models, the implications for NAVER’s cloud-based robotics platform are profound. It would enable the deployment of robots that can handle diverse tasks—from delivering packages to performing delicate maintenance—with minimal human intervention.