In this work, we present a motion planning framework for automated vehicles to drive safely through intersections despite occlusions and the uncertain behavior of the surrounding vehicles. A context representation based on probably-free gaps is proposed as a means to provide, in occluded scenes, richer information to the motion planner compared to representations only based on observed objects. Our solution builds upon the path-velocity decomposition approach. Path planning is performed with state-of-the-art techniques, while a novel trajectory abstraction is used to reason about speed profiles without explicitly generating sequences of accelerations. A reachability-based analysis is as well formulated to efficiently identify the best safe speed profile for every path candidate.The proposed planning workflow is evaluated in roundabout scenarios. Our simulation study shows that the proposed context representation facilitates the decision-making in occluded scenes and that the reachability-based planning strategy is robust, computationally efficient, and outperforms a simpler reactive strategy.