Nano-drones are well suited to indoor object-search tasks because their small size allows them to operate in confined spaces and acquire visual observations from viewpoints that fixed sensors or ground robots may not reach. Approaches that rely exclusively on object detection are not optimal for exploration, as they do not account for the fact that motion planning influences future observations. Instead, since each movement directly determines the next observation, search performance can be improved by exploiting visual evidence also for trajectory generation and action selection. This thesis studies this interaction in a robotic simulation environment with a downward-facing camera, using a physical Crazyflie drone model implemented in Webots.