1. Why we focus on specific use cases of Physical AI (VLA)?
Recent Vision-Language-Action (VLA) models allow robots to look at a scene, understand an instruction, and generate actions. Many efforts aim at general-purpose robots that can do any task, and this is also our long-term goal. However, general tasks demonstrated in labs or homes rarely survive contact with real workplaces. When we move robots into a real workplace, two questions follow: “Where can we get the data that this workplace needs?” and “How can we trust what the robot will do?”
So instead of general tasks, we focus on specific use cases where these questions can be answered. We choose fields where people repeat the same kinds of tasks every day and where the environment and procedures are well defined. Each use case gives us real data, clear success criteria, and a path to deployment, and step by step brings us closer to general robots.
2. What are we working on?
2.1. Physical AI (VLA) for operation assist
Operating rooms are labor-intensive, and surgeons and assistants carry a heavy physical and cognitive load. We are building Physical AI that can safely assist with repetitive surgical tasks, together with our collaborators at Samsung Medical Center (SMC). Our focus is on collecting the data that surgical robots need, and on making robot actions that clinicians can understand and trust.
2.2. Physical AI for lab automation
Many lab experiments consist of repetitive manual steps such as handling samples, pipetting, and operating instruments. We are building Physical AI that can automate these routine lab tasks, so that researchers can spend their time on science rather than on repetitive manual work.
2.3. Physical AI for hotel room housekeeping
Housekeeping is one of the most labor-intensive jobs in hotel operation. Unlike a lab, a hotel room changes every time a guest uses it, and rooms are designed for people, not robots. Together with an industry partner, we are working on robots that can keep up with changing spaces.
More details will be shared as results are published.