vla-jepa-openarm-sbint-step30000

VLA-JEPA policy fine-tuned on SB Intuitions OpenArm SBINT pick-and-place data (LeRobot v3.0), step 30000.

• Train job: train_4306879
• Dataset: sbint_openarm (3 cameras, no rename_map)
• Final train loss ≈ 0.13 (action ≈ 0.01, wm ≈ 0.12)