谷歌DeepMind近日在机器人领域的一系列动作引起了科技界的广泛关注。谷歌不仅在人工智能领域有着深厚的技术积累，此次更是将AI技术应用于机器人身上，通过其最新发布的Gemini 1.5 Pro大模型，机器人不仅能够轻松听从人类指令，进行视觉导览，更能够利用常识推理在三维空间中寻找路径，展现出令人惊艳的智能水平。
Gemini 1.5 Pro大模型的上下文长度达到了百万级，这一特性为机器人提供了强大的环境记忆能力。在真实的工作环境中，工程师们引导机器人进行特定区域的游览，并标记出需要记忆的关键地点，如“刘易斯的办公桌”或“临时办公桌区域”。通过这一系列的操作，机器人在完成一圈游览后，便能够根据其记忆引导人们前往这些标记的地点，即使描述不够精确，机器人也能通过理解指令的意图，找到对应的位置。
这项技术的背后，是谷歌为机器人开发的导航策略——Mobility VLA（Multimodal Instruction Navigation with Long-Context VLMs and Topological Graphs）。这一策略结合了多模态指令导航、长上下文视觉语言模型（VLMs）和拓扑图，使得机器人能够更好地理解并执行复杂的导航任务，实现了与人类指令的高效交互。
Gemini 1.5 Pro大模型与Mobility VLA导航策略的结合，不仅展示了AI在机器人领域的巨大潜力，也为未来的智能办公环境提供了无限可能。未来，随着技术的进一步发展，我们有理由期待机器人在更多场景中展现出更强大的功能，为人类的生活和工作带来更加便捷和高效的体验。
英语如下：
Title: “Google’s Gemini 1.5 Pro Large Model: A New Era of Robot Intelligent Navigation”
Keywords: Gemini Model, Robot Navigation, Contextual Memory
News Content: Google DeepMind’s recent maneuvers in the robotics field have garnered significant attention within the tech community. Not only does Google boast a profound technical foundation in artificial intelligence, it has now innovatively applied AI technology to robots. Through its recently released Gemini 1.5 Pro large model, robots are not only capable of effortlessly following human commands for visual navigation, but also leverage common sense reasoning to find paths in three-dimensional space, showcasing an astonishing level of intelligence.
The Gemini 1.5 Pro large model boasts a context length of millions, a feature that empowers robots with powerful environmental memory capabilities. In real-world operational settings, engineers guide the robots through specific areas, marking out key locations that require memory, such as “Lewis’ desk” or “the temporary office area.” Through these operations, the robot, upon completing a circuit, can guide individuals to these marked locations based on its memory, even when the instructions are not precisely detailed. The robot is able to find the corresponding positions by understanding the intent behind the commands.
The technology behind this is Google’s navigation strategy for robots, known as Mobility VLA (Multimodal Instruction Navigation with Long-Context VLMs and Topological Graphs). This strategy combines multimodal instruction navigation, long-context visual language models (VLMs), and topological graphs, enabling robots to better understand and execute complex navigation tasks and achieve efficient interaction with human commands.
The integration of the Gemini 1.5 Pro large model with the Mobility VLA navigation strategy not only highlights the immense potential of AI in the robotics field, but also paves the way for endless possibilities in future smart office environments. Looking ahead, with the further development of technology, we have every reason to anticipate that robots will exhibit even more powerful functions in various scenarios, bringing about more convenient and efficient experiences for human life and work.
【来源】https://www.jiqizhixin.com/articles/2024-07-15-4
Views: 2