Junaopanshi releases Cog-WM 1.0, a brain-inspired cognitive world model for embodied AI

At the 2026 Pujiang Innovation Forum's brain-inspired embodied intelligence subforum, Junaopanshi Technology released Cog-WM 1.0, a cognitive world model for embodied AI. The company says it combines brain-like latent-space prediction with a JEPA architecture, allowing robots to navigate and manipulate without prebuilt maps.

On the HM3D-ObjectNav subset, reported navigation success rose from 78.50% to 86.89%, and on manipulation benchmarks it outperformed π0.5, which was pretrained on large-scale data, by up to 16% across three tasks. The model has been validated on a quadruped robot and a wheeled humanoid, and its navigation ability is already used at customer inspection and patrol sites; the method is described in an arXiv paper.

Original sources (Chinese)