🤖 Hy-Embodied-0.5-VLA-UMI — Dual-Arm Action Trajectory Demo
A Vision-Language-Action model from Tencent Hunyuan (5B params, flow-matching action expert). Give it three camera views (top / left-hand / right-hand) and a language instruction; it predicts a 50-step dual-arm end-effector action chunk (10-DOF per arm: xyz + rot6d + gripper), visualized in 3D below.
Examples