Frontier AI Advances Autonomous Physical Science and Embodied Agents

Google DeepMind's Gemini Co-Scientist reportedly controlled laboratory equipment, generated machine code, and iterated physical experiments, including 2D-material growth and safer synthesis exploration. NavMCP results indicate that navigation execution, evidence-seeking, and persistent memory can improve long-horizon physical-agent performance without retraining, though reproducibility, safety, and real-world robustness remain open questions.