In 2023, Om AI achieved an unexpected breakthrough when a multimodal model never trained on surveillance footage outperformed veteran small models fine-tuned on surveillance data for years in understanding monitoring scenarios, confirming for CEO and Chief Scientist Dr. Zhao Tiancheng that "multimodal training can bring generalization capabilities to the open physical world." Three years later, this model evolved into VLX, the world's first end-side streaming multimodal model series dedicated to physical AI, which pioneers a brand-new "end-side native streaming multimodal" architecture designed from Day 1 for end devices. The VLX model is the first to close the full loop of physical AI on end-side devices: "continuous perception + precise positioning + action decision-making". As of the summer of 2026, the AI industry's focus has shifted from digital AI to physical AI, but the industry has not converged on a unified path, with multiple technical routes coexisting. While the industry debates between VLA and world models, Om AI's VLX series has completed the commercial closed loop of physical AI from simulation experiments to industrial deployment, equipping robots, drones, wearable devices, security cameras, AI PCs and other physical terminals to achieve the leap from "passively executing commands" to "actively adapting to scenarios". Zhao Tiancheng, who graduated with a doctorate from Carnegie Mellon University's Language Technologies Institute in 2019, chose this path despite easily accessible positions as a professor or researcher.