Imagine a disaster scenario: a chemical spill unfolds, and a first responder needs to deploy a robot for triage. The robot receives a simple command: "Triage impacted regions." But how can a robot understand such an incomplete instruction, especially in an unstructured, unfamiliar environment? Researchers are tackling this challenge with innovative semantic planning techniques. Their work focuses on enabling robots to infer missing information and explore dynamically, even with minimal initial guidance. The key lies in combining the power of Large Language Models (LLMs) with a real-time understanding of the robot's surroundings. The system works by first interpreting the high-level mission command using an LLM. The LLM breaks down the complex task into smaller, actionable steps, such as identifying relevant semantics (like "victims" or "damaged regions") and exploration targets. Crucially, the robot isn’t limited by pre-defined maps; it actively explores and updates its understanding of the environment, refining its plan as it gathers new information. Safety is paramount in these scenarios. Therefore, a validation module continuously checks the LLM's proposed actions, ensuring they are both logically and spatially sound. This prevents the robot from attempting unsafe or impossible maneuvers, even in dynamic, unpredictable environments. Real-time feedback from sensors allows the robot to adapt to unexpected obstacles and refine its plan as needed. The system has been tested in both simulated and real-world settings, demonstrating promising results. In large-scale outdoor environments, robots were able to successfully complete missions involving semantic route inspection, multi-object search, and even air-ground teaming with UAVs. These experiments showcase the potential of this technology to revolutionize how robots operate in complex, real-world missions. While challenges remain, this research marks a significant step towards more autonomous and adaptable robots. Future research could focus on improving the efficiency of LLM reasoning and extending these techniques to multi-robot collaborations. Imagine a future where robots can understand and execute complex tasks with minimal human intervention, effectively becoming indispensable partners in disaster response and other challenging domains.