JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language Models