StradVision, a developer of automotive artificial intelligence object recognition software, plans to expand the AI vision technology it has built up in autonomous driving into physical AI fields such as robotics, defense and infrastructure. Its first target is autonomous mobile robots, or AMRs, which navigate on their own at airports and logistics facilities, and the company is already discussing possible applications with potential customers. Over the longer term, it aims to widen the reach of its recognition software to aviation and drones as the automotive technology advances.
Kwon Tae-san, StradVision's chief operating officer, met with The Seoul Economic Daily recently at the company's Dongtan office in Hwaseong, Gyeonggi Province. "In physical AI as well, you ultimately need recognition technology that plays the role of the human eye," he said. "We are focusing first on the AMR field and holding discussions with customers there, and going forward we should be able to widen the scope of application to a broader human world."
The medium- to long-term technology road map StradVision has laid out also points beyond the car. The plan is to concentrate on developing automotive software in its current product lineup — Front Vision, Surround Vision and Multi Vision — and then, over the longer run, direct some development resources to adjacent industries. Defense, infrastructure, robotics, aviation and drones are cited as next-generation applications.
Robots, in particular, hinge on the same core capability as self-driving cars: recognizing the physical world. To move and perform tasks on their own, they must identify people, objects and obstacles through cameras and other sensors and grasp their own position and surroundings. That is why StradVision believes the object recognition technology it has accumulated for cars can be extended to physical AI. Independence from specific hardware is another technical strength the company points to. Because it has not locked its software to a particular chip or camera combination, it now supports more than 30 types of automotive chipsets.
Data is the other main pillar of the physical AI expansion. StradVision plans to commercialize its own data pipeline, SV DataFlow, starting next year and offer it to outside customers. The move turns data preprocessing, conversion and labeling technology that it has used internally as a development tool for autonomous driving software into a separate software-as-a-service business.
"One consideration was that starting this business would make expansion into the physical AI area easier," Kwon said. "In the end, demand for an automated data pipeline will arise in any industry that needs vision recognition, including physical AI." He said the company would widen its business reach into the physical AI market by combining the AI "eye" it has built in the mass-production automotive market with the data technology that continuously trains it.
