This might be for improving their camera-based vision system by collecting both camera inputs and accurate ground-truth data via LIDAR to train the system with.