Imagine a team of robots working together seamlessly, not through complex pre-programmed routines, but through the power of language. This is the vision brought to life by researchers exploring how Large Language Models (LLMs), like the ones powering chatbots and AI assistants, can revolutionize multi-robot collaboration. Traditionally, coordinating multiple robots, especially ones with different abilities (like a drone, a robotic dog, and a robotic arm), has been an incredibly challenging task. Pre-programmed instructions often struggle to adapt to unexpected situations, making complex collaborations in dynamic environments nearly impossible. This new research introduces COHERENT, a groundbreaking framework that allows heterogeneous robots to work together by using LLMs to communicate and coordinate actions in real-time. The system uses a clever 'Proposal-Execution-Feedback-Adjustment' cycle. A central LLM acts as a task assigner, breaking down complex instructions (like "move the apple from the sofa to the dining table") into smaller subtasks for each robot. The robots then attempt to execute these subtasks, providing feedback to the central LLM about their success or any difficulties encountered. This feedback loop allows the system to adapt on-the-fly, adjusting the plan as needed. The researchers tested COHERENT in a variety of simulated environments, including apartments, a grocery store, and a restaurant, with tasks that required increasingly complex collaboration. The results? A significant improvement in success rates and efficiency compared to traditional methods. What's even more exciting is that COHERENT isn't just a theoretical concept. The team demonstrated its real-world potential using a physical robot team, showcasing a future where robots can understand and execute complex instructions, collaborate dynamically, and adapt to real-world challenges with remarkable flexibility. This research opens doors to a future where teams of specialized robots could tackle complex tasks in various fields, from disaster relief and manufacturing to healthcare and exploration. While still in its early stages, this approach hints at a transformation in how we design and deploy robotic systems, using the power of language to unlock unprecedented levels of collaboration and adaptability.