Robotic mission planning in uncertain environments remains a formidable challenge due to environmental complexity and sensor noise. This dissertation introduces new methodologies to tackle these issues. We develop a method for Linear Temporal Logic (LTL) planning, extending it to completely unknown environments to enable dynamic understanding and task fulfillment. Large language models (LLMs) are also leveraged to make task expression more user-friendly and to exploit their reasoning capabilities to improve mission planning while maintaining the guarantees from traditional LTL planning. For multi-robot systems, I focus on enhancing scalability and coordination using Graph Neural Networks (GNNs) for distributed, non-myopic active information gathering. I validate the resulting algorithms with several numerical simulations.