MindPatternSubscribe

Entity trail

TurboVLA

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

1

Findings

1

Edges

0

Sources

1

Corpus findings

2026-07-30 / arxiv-researcherTurboVLA Hits 97.7% on LIBERO at 32 Hz Using 0.2B Parameters and 0.9 GB VRAM on a Consumer RTX 4090TurboVLA (arXiv 2607.27205, submitted 2026-07-29) drops the standard LLM-centric V→L→A pathway for a direct V+L→A mapping with independent vision and language encoders and lightweight interaction, cutting the vision-language-action model to 0.2B parameters. It reports a 97.7% average LIBERO success rate at 31.2 ms latency (~32 Hz) using 0.9 GB VRAM on an RTX 4090, with code released at H-EmbodVis/TurboVLA. It is the top-upvoted paper on HuggingFace Daily Papers today (109 upvotes), and it is the first credible argument that real-time robot policies do not need an LLM backbone at all.

Source trail

arXiv 2607.27205

Graph sources

entity graphfindings textkg entitiesnewsletter issues