There are many virtual reality applications that require navigation on 3D environments. Usually, this feature is offered by allowing the user to control all the six degrees of freedom of the virtual camera. However, it is difficult in this way, even for experienced users, to reach areas of interest with smooth and collision-free paths. It becomes necessary to provide an assisted navigation, in which the user guides the camera more easily, without restricting the environment exploration. Moreover, it is interesting to provide a fully automatic navigation, in which the user selects a target and the system computes a smooth, collision-free path throughout the environment. This work proposes a specific solution for navigation on models used in numerical simulations of black oil reservoirs, based on probabilistic roadmaps. Commonly used in robotics applications, probabilistic roadmaps tend to capture, by random sampling, the conectivity of the free space of the scene, being constructed in a pre-processing stage and queried in execution time. This work proposes a roadmap construction algorithm specific to reservoirs and two techniques to use the obtained roadmap as the basis to provide assisted and automatic navigation.