KO EN

CAT-01 — Physical AI Models & Core Tech

World Models · VLA · Foundation Models · Training

CAT-01 — Physical AI Models & Core Tech examines the foundation model stack enabling embodied agents to perceive, simulate, and act within 3D physical environments.

Core Research Track

  1. [Pillar] Physical AI Architecture Overview: From Generative LLMs to Embodied Intelligence (LLM / VLM / VLA / WFM 4-Layer Stack)
  2. How World Models Work: Latent space physics prediction and the NVIDIA Cosmos WFM lineage
  3. Vision-Language-Action (VLA) Pipeline Analysis: Translating multimodal tokens into continuous robot control (Figure Helix-02, Isaac GR00T N1.7, OpenVLA)
  4. Foundation Policy Training: Teleoperation imitation learning and Diffusion Policy optimization