A multimodal learning framework that maps visual observations and natural language instructions to robot motor commands.