aiqbench analysis
This paper proposes a training-free way to fix a known weakness in 3D language maps: target objects that vision-language distillation often fails to locate. It blends each target region's visual embedding with its text embedding, using a closed-form coefficient chosen to be as small as possible while still making the target win. Readers should note the reported gains, including roughly doubled mIoU on Replica, are benchmark results, and the approach's value for robotics depends on whether it holds up beyond those three benchmarks and the single legged-robot demonstration.
Training-Free Target-Aware Refinement of 3D Gaussian Language Map
arXiv:2610.21619v1 Announce Type: new Abstract: Distilling vision-language features into 3D Gaussian Splatting yields a language map that can be queried with arbitrary text. In deployment, however, a small set of user-specified targets must be reliably located, which current distilled maps often miss, while open-vocabulary queryability for other words must remain intact. We propose a training-free, target-aware refinement of an already-distilled language map. Guided by a text-promptable segmenter, the visual embedding inside each detected target region is blended with the target text embedding, a convex combination whose coefficient sets the amount of injected prior, while non-target regions are left untouched. Injecting too much overwrites appearance cues and general queryability, so we seek, for each target, the smallest coefficient that still lets it win, and we obtain every coefficient in closed form, leaving no hyperparameters to tune. We further identify a failure mode of injection in language-goal localization. Text embeddings share a common component, and injecting them makes the standard cosine read-out query-agnostic, which a centering step in the read-out removes. Across three benchmarks, the refinement consistently improves open-vocabulary segmentation, roughly doubling the base mIoU on Replica, and localizes language goals more reliably than the base map. It also transfers as a plug-in to off-the-shelf field-based and object-level language maps and supports a real-world demonstration in which a legged robot navigates to natural-language goals. Project page and code: https://lab-of-ai-and-robotics.github.io/target-aware-langmap/
Abstract:Distilling vision-language features into 3D Gaussian Splatting yields a language map that can be queried with arbitrary text. In deployment, however, a small set of user-specified targets must be reliably located, which current distilled maps often miss, while open-vocabulary queryability for other words must remain intact. We propose a training-free, target-aware refinement of an already-distilled language map. Guided by a text-promptable segmenter, the visual embedding inside each detected target region is blended with the target text embedding, a convex combination whose coefficient sets the amount of injected prior, while non-target regions are left untouched. Injecting too much overwrites appearance cues and general queryability, so we seek, for each target, the smallest coefficient that still lets it win, and we obtain every coefficient in closed form, leaving no hyperparameters to tune. We further identify a failure mode of injection in language-goal localization. Text embeddings share a common component, and injecting them makes the standard cosine read-out query-agnostic, which a centering step in the read-out removes. Across three benchmarks, the refinement consistently improves open-vocabulary segmentation, roughly doubling the base mIoU on Replica, and localizes language goals more reliably than the base map. It also transfers as a plug-in to off-the-shelf field-based and object-level language maps and supports a real-world demonstration in which a legged robot navigates to natural-language goals. Project page and code: this https URL
8 pages, 6 figures. Project page: this https URL
Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2610.21619 [cs.RO]
(or arXiv:2610.21619v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2610.21619 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 332 words
Related coverage
[AINews] AI Engineer NYC: Gemini 4 Argon, Vals AI Benchmarks
Latent Space · Oct 14, 2026
LFM: Leveraging Foundation Models for Source-Free Universal Domain Adaptation
arXiv · Jul 21, 2026
[Paper] Distillation for Incrimination and Distillation for Capabilities
LessWrong · Oct 9, 2026
[AINews] not much happened today
Latent Space · Sep 10, 2026