Surveillance, search and rescue, or monitoring missions in hazardous environments can be executed by a team of heterogeneous robots embedding autonomous capabilities to detect activity or intruders, to collect samples or to disarm traps. For those applications, one can deploy a team of ground and aerial robots, equipped with specific sensors and autonomous navigation and physical interaction capabilities. The embedded robots capabilities are arranged in a 3-layer based architecture comprising decisional, executive and function layers, and in particular, implementing a skill-based layer as the executive layer including fault recovery to ensure robust behaviors. However, embedded decision-making becomes challenging in uncertain environments with partial (or to-be-discovered) information about locations to visit or terrain traversability. To address it, non-deterministic planning formalisms are well adapted, such as FOND or contingent planning approaches, but at the cost of a solution plan growing exponentially as multiple actions’ outcomes increase branchings –- leading sometimes to plans that are unreadable to human operators. In this paper, we therefore present a formal and automatic factorization method of such plans, in order to improve readability to users, as well as reduce the cost of formal verification of robotics behavior before deployment. Additionally, we present how this factorized plan can be translated to robust hierarchical skill compositions for the executive layer and therefore enhancing the reliability of planned actions with recovery behaviors.