TARS AWE 3.5 aims for generalization with embodied-native architecture

FellMentKE · x · 2026-08-18

Real robotics progress lies in understanding why actions work and transferring that understanding. TARS AWE 3.5 features an embodied-native architecture, latent action & geometry priors, and world-engine-driven post-training, forming a closed loop designed for generalization across diverse tasks and environments.

Original post →

More from Embodied

Embodied channel →