LIVIN
LIVIN: Benchmarking Spatial and Embodied Intelligence in Digital Twins of Lived-In Homes
LIVIN evaluates spatial and embodied intelligence using digital twins of 30 lived-in homes, covering 3D detection, reconstruction, navigation, and loco-manipulation tasks.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info livin
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
4 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-6 Astra | 41.5 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| DeepSeek V4.1 Flash | 11.4 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| OmniNav | 6.7 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| DualVLN | 4.4 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
LIVIN comprises digital twins of 30 diverse lived-in homes with 319 rooms and 12,808 object instances, constructed via a human-in-the-loop workflow preserving realistic layouts and articulated objects.
It benchmarks 3D detection, reconstruction, navigation, and loco-manipulation, revealing that current methods struggle with dense clutter, occlusions, and constrained interaction regions in realistic domestic environments.