# KRVF: A Source-Aware Semantic Voxel World Representation for Edge Mobile Manipulation

arXiv: https://arxiv.org/abs/2606.26321 · HTML: https://mlsanity.com/paper/2606.26321/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Embodied and robot world models, one of 11 directions in World models.
- Builds on 1 of its 12 references that are in the map.
- First posted on arXiv 24 Jun 2026.

## Placed in

- World models / [Embodied and robot world models](https://mlsanity.com/wm/embodied-and-robot-world-models/)

## Builds on (most cited first, 1 of 1)

- [YOLOv3: An Incremental Improvement](https://mlsanity.com/paper/1804.02767/) (2018)

## Abstract (arXiv)

Mobile manipulators need world models that are current, queryable, semantically meaningful, and usable under edge-compute constraints. This technical report presents KRVF, a source-aware semantic voxel world representation for edge mobile manipulation. Unlike reconstruction-centric mapping pipelines that primarily optimize global geometric fidelity, KRVF represents local world state as task-oriented voxels that encode occupancy, color, semantic evidence, temporal freshness, and evidence source. The representation separates measured occupancy from semantic-prior hypotheses, enabling depth-failure-aware object reasoning without silently corrupting persistent geometry. KRVF also closes a feedback loop between mapping and sensing by rendering map-prior depth for repair, and exposes task-level query operators for semantic objects and grasp candidates. The report formalizes the KRVF representation and documents a ROS 2 implementation that turns online RGB-D observations into a task-facing robot memory.