Imagine robots seamlessly navigating the chaotic aftermath of a disaster, or effortlessly assembling complex structures on a bustling construction site. This isn't science fiction, but a future rapidly approaching thanks to groundbreaking advancements in AI, particularly in the realm of foundation models. These powerful AI systems, trained on vast amounts of data, are revolutionizing how robots perceive, reason, and act in unstructured environments, places characterized by unpredictable changes and limited information. Unlike traditional robots programmed for specific tasks, foundation model-equipped robots can generalize their knowledge, adapting to new situations and performing diverse tasks with minimal human intervention. Recent research has focused on leveraging Large Language Models (LLMs) like GPT-3 and -4, and visual models like CLIP, to enhance robot capabilities in understanding human instructions, perceiving complex scenes, and even making decisions in uncertain scenarios. For example, in construction, foundation models are being used to improve safety training, recognize hazards, and automate project management tasks. In disaster zones, they’re proving invaluable for analyzing social media data to extract real-time information and automate disaster reporting. However, the journey towards fully autonomous robots in unstructured environments is not without its challenges. A key hurdle is the scarcity of 3D data needed to train robots for spatial reasoning and interaction with the physical world. Researchers are actively working on solutions like personalized Digital Twins, virtual environments, and powerful simulators to overcome this limitation. Another critical issue is the occasional "hallucination" of foundation models, where they generate incorrect or illogical outputs. To mitigate this, researchers are exploring methods like Retrieval-Augmented Generation (RAG) and developing systems that allow robots to seek human assistance when uncertain. Latency, the delay in processing information, is another challenge, especially in time-sensitive situations. The emergence of faster internet connections and more efficient edge computing solutions offers promising solutions. Finally, improving human-robot interaction is crucial. Future research aims to develop robots capable of genuine interaction, empathy, and ethical decision-making, fostering trust between humans and machines. While current research primarily focuses on cognitive tasks, such as perception and planning, future breakthroughs will focus on enhancing robots' physical acting abilities—control, manipulation, and navigation—to truly unlock their potential in unstructured worlds. The path towards full robot autonomy involves exciting explorations of multi-agent systems, where specialized AI agents collaborate within a single robot, and novel robot collaboration structures, where robots learn and adapt together. The future of robots in unstructured environments is bright, promising transformative changes in how we interact with and shape the world around us.