Autonomous systems makes decision based on information received from various sensors. The difficulty of relying on sensor outputs is that they don't always produce correct observations and at times don't produce any observations at all. To improve the level of safety of autonomous systems we need to be able to understand under which conditions the sensors can be trusted to behave reliably. To this end we propose learning sensor monitors. Furthermore to support safety, we propose deploying shielding - a method of disallowing an autonomous agent from making unsafe decisions. In particular we are interested in robust shields, which can operate under condition when some of the sensor data is deemed unreliable or unavailable. In this project, we aim to design a framework, which combines contextual sensor monitors, a RL (reinforcement learning) agent and an array of learned shields. The result is a framework which can at runtime asses reliability of sensor outputs and use this information to enforce safe behavior.