Summary:KAIST’s VOTP technology enables physical AI to learn human judgment from just a few videos, reducing development costs and paving the way for robots and autonomous systems that act in human-aligned ways.

June 10, 2026 — Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have developed a breakthrough technology enabling physical AI systems to learn human judgment criteria from just a few videos. The method, called VOTP (Video-based Optimal TransPort Preference), allows AI to infer human intentions and preferences without requiring thousands of manually evaluated data points.

VOTP overview diagram. Credit: The Korea Advanced Institute of Science and Technology (KAIST)

Physical AI—encompassing autonomous robots, surgical machines, and self-driving vehicles—requires systems to evaluate and select actions that align with human expectations. Traditionally, building such reward functions demanded extensive human feedback, making development costly and time-consuming. VOTP addresses this challenge by mimicking the way humans learn from limited demonstrations. By analyzing a small set of “good” and “bad” action videos, AI can generalize human-preferred behavior across diverse tasks and environments.

Extensive testing confirmed that VOTP significantly reduces the need for large-scale human evaluation while maintaining accurate judgment and generalization. The approach is applicable across industrial robotics, humanoid robots, drones, autonomous vehicles, surgical robots, and AI agents operating in digital environments. By lowering development costs and accelerating learning, VOTP represents a foundational technology for practical, human-aligned physical AI.

Professor Chang D. Yoo, who led the research, emphasized that understanding human intentions is central to physical AI, stating that “VOTP is a core technology that will accelerate the era of robots making human-like judgments.”