This paper presents results from an Early Stage Innovations project entitled “Deep Contact Graph Routing for Lunar Operations,” which aims to use machine learning to improve wireless communication in lunar environments. This paper considers a rover traveling between two distant waypoints while communicating with a base station. The rover accounts for wireless connection strength as it plans its path between waypoints. While this would be simple with a known base station location, allowing the transmitter to be in an unknown location yields a solution with a variety of applications. To evaluate trajectories in terms of connection strength, we augment a Model Predictive Control (MPC) local planner with a connection strength heatmap generated using the signal’s Angle of Arrival (AoA), which is extracted from Channel State Information (CSI), as a proxy for connection strength. This system is modular and enables easy experimentation with alternative heatmap generation methods, such as multiple rover data fusion and deep neural networks. Beyond efficient data transfer, our approach has a variety of applications, ranging from multi-rover cooperative exploration to lunar search-and-rescue of rovers that have lost satellite communication.