Underwater VLA policies beat baseline; ACT beats SmolVLA at sea
A 14-model DynamicVLA study and an ROV case study at a Malta wreck also open-source the BLUEMBODIED KIT.
42
AI model
SmolVLA
- Mentions · 7d1
- Mentions · 30d4
- All mentions4
- First seenOct 17, 2026
- Last mentioned3d ago
Timeline
4 articles · newest firstBacktrack lets frozen vision-language-action policies retry failed motions
On LIBERO-Long, failure rates fall 19–48% and success rises by up to 7.4 points; the head adds 0.48–0.88% parameters.
42
Activation study finds VLA motor programs ignore shifted scenes
Injected task-A activations made pi0.5 replay that trajectory in 99.6% of 1,968 episodes.
52
VLA-Rolling reuses action tails to cut VLA denoising steps
Across 16 frozen model-benchmark pairs it lowered chunk inference cost, with up to 1.48 times speedup and higher success in 14.
42
Appears with
- pi0.5ai model · 2
- X-VLAai model · 1
- BLUEMBODIED KITproduct · 1
- LIBEROproduct · 1
- π0.5ai model · 1
- π0-FASTai model · 1
- DynamicVLAai model · 1
- Action Atlasproduct · 1
- UR10eproduct · 1
- MetaWorldproduct · 1
- MolmoAct2ai model · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.