Edge-assisted vehicle-to-everything (V2X) motion planning is an emerging paradigm to achieve safe and efficient autonomous driving, since it leverages the global position information shared among multiple vehicles. However, due to the imperfect channel state information (CSI), the position information of vehicles may become outdated and inaccurate. Conventional methods ignoring the communication delays could severely jeopardize driving safety. To fill this gap, this paper proposes a robust V2X motion planning policy that adapts between competitive driving under a low communication delay and conservative driving under a high communication delay, and guarantees small communication delays at key waypoints via power control. This is achieved by integrating the vehicle mobility and communication delay models and solving a joint design of motion planning and power control problem via the block coordinate descent framework. Simulation results show that the proposed driving policy achieves the smallest collision ratio compared with other benchmark policies.