RoCNet++ is a novel method for accurate and robust point cloud registration, introducing a triangle-based descriptor and a Farthest Sampling-guided Registration (FSR) module. The triangle-based descriptor captures local geometric properties by encoding the angles of triangles formed by each point and its nearest neighbors, ensuring invariance under rigid transformations. This descriptor is integrated into the RoCNet architecture to estimate correspondences between point clouds. The FSR module estimates multiple rigid transformations by sampling points based on farthest point exploration, selecting the transformation with the highest number of inliers. RoCNet++ is evaluated on ModelNet40, KITTI, and 3DMatch datasets, showing improved performance in terms of accuracy and robustness, especially in noisy and partial data scenarios. The method outperforms existing approaches in registration metrics, achieving high precision, accuracy, and recall. The triangle-based descriptor and FSR module are also applicable to other registration algorithms. The results demonstrate that RoCNet++ provides accurate and efficient point cloud registration across various conditions.RoCNet++ is a novel method for accurate and robust point cloud registration, introducing a triangle-based descriptor and a Farthest Sampling-guided Registration (FSR) module. The triangle-based descriptor captures local geometric properties by encoding the angles of triangles formed by each point and its nearest neighbors, ensuring invariance under rigid transformations. This descriptor is integrated into the RoCNet architecture to estimate correspondences between point clouds. The FSR module estimates multiple rigid transformations by sampling points based on farthest point exploration, selecting the transformation with the highest number of inliers. RoCNet++ is evaluated on ModelNet40, KITTI, and 3DMatch datasets, showing improved performance in terms of accuracy and robustness, especially in noisy and partial data scenarios. The method outperforms existing approaches in registration metrics, achieving high precision, accuracy, and recall. The triangle-based descriptor and FSR module are also applicable to other registration algorithms. The results demonstrate that RoCNet++ provides accurate and efficient point cloud registration across various conditions.