Autonomous lunar rovers face a unique perception challenge near the poles, where low sun angles create long, high-contrast shadows that can break stereo vision and terrain segmentation. Gathering enough real sensor data to train and test algorithms for these conditions is expensive and limited. The new open-source MARE simulator aims to fill that gap by generating large amounts of synthetic sensor data with accurate ground truth.
Built on Unreal Engine 5 and ROS 2, MARE has three integrated parts: Terrain Studio, which creates cratered and rocky digital elevation maps; a rover project with physics-based models and plugins for stereo, lidar, event, and SPAD cameras; and a guidance, navigation, and control stack with web-based mission control. The simulator streams time-stamped sensor data alongside pixel-accurate ground truth, allowing researchers to validate perception pipelines under repeatable, extreme lighting conditions.
The authors report a mean Intersection over Union of 0.79 between predicted and rendered shadows at low sun elevations, and stereo depth analysis confirms that shadowed regions have reduced pixel coverage compared to sunlit terrain. These results suggest MARE can realistically reproduce the visual challenges of polar lunar exploration, offering a low-cost complement to field campaigns and analog facilities for developing robust rover autonomy.