ReaLiTy and LADS: A Unified Framework and Dataset Suite for LiDAR Adaptation Across Sensors and Adverse Weather Conditions

ReaLiTy addresses a critical bottleneck in autonomous driving (AD): the lack of paired, multi-sensor, multi-weather data for LiDAR perception. While generic data augmentation exists, 'physics-informed' transformation—modeling beam divergence, atmospheric attenuation, and sensor-specific intensity returns—is a deep-tech moat. The 0-star count is misleading due to its age (6 days), but the 4 forks indicate immediate peer interest from researchers. The project's defensibility lies in its specialized physics kernels which are harder to replicate than standard GAN/Diffusion wrappers. However, platform risk is high because NVIDIA (via Omniverse/DriveSim) or Waymo could integrate similar physics-informed sensor simulation natively into their stacks. The primary value is as a standardized benchmark (LADS) for domain adaptation, which creates 'data gravity' if researchers adopt it for comparison. Its long-term risk is the rise of end-to-end generative world models (like Sora for LiDAR) which might eventually learn these physics implicitly without explicit modeling.