Robo2VLM: Improving Visual Question Answering using Large-Scale Robot Manipulation Data | Read Paper on Bytez