The arXiv announcement for VEOcc++ describes a method for semantic occupancy mapping, a task that supports embodied perception both during active exploration and in later scene analysis. The paper frames these as complementary requirements: an online mode that maintains an up-to-date global map incrementally, and an offline mode that allows for more detailed refinement after exploration.
To address both, VEOcc++ learns persistent voxel feature maps, which appear to be the central mechanism for bridging the online and offline stages. The abstract emphasizes this representation as the key contribution, though the full technical details are not included in the available text.
Since only a single source was provided, there is no independent comparison to confirm or contrast the claims. The summary here reflects only what the arXiv abstract states.