3D-AEIR:3D Annotation for Embodied Intelligent Robots

The realization of Artificial General Intelligence (AGI) relies on the advancement of Embodied AI, and embodied perception is one of the most representative technologies in Embodied AI. One of the key challenges faced by this technology is the difficulty in obtaining high-quality training data, primarily because of the complex nature of the application scenes for embodied intelligent robots and the lack of annotation methods capable of adapting to such varied environments. We propose 3D-AEIR——an annotation method based on multi-sensor fusion and image segmentation, as a solution to this challenge, with the aim of improving the efficiency and accuracy of annotation 3D objects in different scenes. The experimental results on both public and internal datasets demonstrate that our method significantly improves the efficiency and accuracy of 3D object annotation, offering a novel solution for dataset creation in diverse scenes.