Adverse weather conditions, such as fog and rain, pose major challenges for autonomous vehicle perception by degrading sensor data quality. However, existing datasets rarely offer reproducible conditions or quantitative evaluations of sensor performance. To address this, we present MuFoRa, a multimodal dataset acquired in a controlled indoor facility using a stereo camera and two solid-state LiDAR sensors. Data were collected under clear and night conditions, as well as under simulated fog (5 m to 160 m visibility) and rain ( \({20}\,\text {mm}\,\text {h}^{-1}\) to \({100}\,\text {mm}\,\text {h}^{-1}\) ), with target distances from 5 m to 50 m. Sensor degradation is assessed using normalised image entropy for camera data and the inlier ratio of points relative to their distance from the surface of a spherical target for LiDAR data. Results show that camera performance is moderately affected by rain and strongly affected by fog, with normalised entropy dropping by 35% and 57%, respectively. LiDAR performance remains stable under moderate rain, but degrades under dense fog and heavy rain, with inlier ratios reduced by up to 41% and 44%, respectively. Additionally, we propose a semi-automated approach for evaluating calibration accuracy using a spherical target, which includes a circle detection benchmark with 24 deep learning models. The MuFoRa dataset enables reproducible quantitative benchmarking of perception and calibration under controlled weather conditions, supporting the development of robust multimodal perception systems for autonomous driving.
