Recent advancements in robotics have significantly enhanced the capabilities of robots in various mission planning tasks such as environmental monitoring, structure inspection, and search and rescue. A successful example is that of industrial robots, programmed to perform assembly operations within predictable, structured environments. In contrast, in unstructured environments like residential houses the robots often struggle with unforeseen circumstances, such as poorly mapped areas or unexpected human interactions, highlighting the limitations of current technology in handling mission planning under uncertainty. This dissertation addresses these challenges by exploring new planning methods for mobile mission planning in uncertain environments. The first part of this dissertation introduces a method for a single robot tasked to rearrange a set of moving objects the locations of which are á-priori unknown while also navigating in an uncertain environment. The robot is equipped with noisy range-sensor and a gripper and the task is expressed with a Linear Temporal Logic (LTL) formula. In this part, we introduce a hybrid control architecture to solve mission planning under these environmental and sensing uncertainties. Under certain assumptions, the method is proven to be probabilistically complete. Next, the complexity of the problem is further scaled up by requiring the robot to navigate in an entirely unknown environment and interact with objects about which it has no prior knowledge of their nature or location. A novel reactive mission planning is introduced where the robot dynamically builds an understanding of its surroundings while adapting its behavior to unforeseen circumstances to fulfill its mission objectives. In the next chapter, the thesis focuses on abstracting task expression by leveraging large language models (LLMs). This chapter makes task specification more user-friendly by allowing tasks to be expressed in natural language and enhances the robot’s ability to comprehend semantic relationship between objects leading to improved and faster mission planning solutions in known environments. Furthermore, this architecture is the first method for mission planning with LLMs that is accompanied by optimality and completeness guarantees. The final part of this dissertation takes the first step towards addressing mission planning under uncertainty with multiple robots. The robots operate in completely known environments and must visit a set of geometric goals their locations of which is á-priori unknown. The approach mainly focuses on designing scalable methods for multi-robot active information acquisition. Two methods are introduced, a distributed non-myopic sampling-based method and a Graph Neural Network (GNN) based method enabling robots to collaboratively explore and gather information in uncertain environments.