OpenGalaxea

Use when the user wants to set up and run a minimal end-to-end LIBERO fine-tuning/training in this GalaxeaVLA (G0.5) repo from scratch — builds the venv, wires the checkpoints + LIBERO LeRobot data + stats into the exact layout the configs expect, then runs a single-GPU smoke to prove the data→model→loss→optimizer loop works. Triggers: "配一个最小 libero 训练", "set up libero training", "跑通 libero finetune", "minimal libero smoke".