The recent advancements in the field of robotic perception and interaction have shown significant progress in enhancing the accuracy and efficiency of human-object contact detection, 3D scanning, semantic mapping, distributed shape learning, and volumetric mapping. Innovations in depth-aware perspective interaction and object texture restoration have led to more precise detection of contact areas, overcoming issues related to occlusions and view blockages. In the realm of 3D robotic scanning, boundary exploration techniques have improved the efficiency of object capture and reconstruction by optimizing view overlaps and camera positioning. Semantic mapping in horticultural environments has been made more efficient through the use of probabilistic maps and novel information utility functions, addressing challenges related to segmentation noise and environmental variability. Distributed learning of complex objects has been advanced by the adoption of Gaussian kernels, enabling networked robots to share functions through finite constraints. Lastly, volumetric mapping has seen improvements in panoptic segmentation quality through kernel density estimation, enhancing the precision of 3D scene reconstruction and object localization. These developments collectively push the boundaries of robotic capabilities in perception, interaction, and mapping, paving the way for more sophisticated and practical applications in various domains.

Noteworthy papers include one that introduces a perspective interaction HOT detector achieving state-of-the-art performance on benchmark datasets, and another that proposes a boundary exploration NBV network for efficient 3D scanning, outperforming existing methods in scanning efficiency and overlap.