The ability to manipulate objects in an unstructured environment is a key capability needed to fulfill the potential of humanoid robots. Proper mobile platform placement in the presence of obstacles is the first constraint that needs to be fulfilled for a successful object manipulation execution. In this paper, we investigated the applicability of neural networks in deciding the mobile platform placement for a given task. Inv-Reach Net architecture was introduced which was used to update the inverse reachability map (IRM - mobile manipulator's capability representation) to account for the obstacles in the environment. The updated IRM was used to calculate an optimal mobile platform pose. Results indicated that in calculating an updated IRM, Inv-Reach Net was significantly faster than the traditional method with minor errors. The increased accuracy of the updated IRM leads to successful optimal mobile platform placement. The Inv-Reach Net may be used to increase the task execution success rate of humanoid robots.