NAT: Scaling Causal Neural Action Tokenizers Across Embodiments
Paper available
Authors & affiliations (3)
Author order: Original order
- First published
- Not provided
- Latest paper revision
- Not provided
- Venue
- Not provided
NAT VLA 450M is a robot policy built on Liquid AI’s LFM2-VL-450M. It turns images, natural-language instructions and robot state into time-ordered NAT action tokens for streaming control. LeRobot supports fine-tuning and token-prefix real-time chunking.
Paper available
Author order: Original order
SourceVerified on · 2026-11-22
Verified on · 2026-11-22
| File | Access | Weights license | Commercial use |
|---|---|---|---|
| NAT VLA 450M Pretrained | Access | Weights license | Commercial use ConditionalTerms Verified on · 2026-11-22 ConditionsLFM Open License v1.0 applies to the policy checkpoint. See section 5 for the annual revenue threshold of USD 10 million and the qualified nonprofit research provision. |
SourceVerified on · 2026-11-22
Fine-tune on target robot data. With the default tokenizer, streaming and token-prefix RTC require at most seven action dimensions; wider actions are decoded in groups and returned as a full chunk. The paper’s real-robot results used a separately fine-tuned Trossen AI Solo policy. Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.
Liquid AI LFM2-VL-450M
Apache-2.0
Cross embodiment
Official
action_control
Actions
Time-ordered NAT action tokens
Robotic arm
Multiple tasks
1,800 hours of VLA pretraining data
BridgeData V2 / DROID
Imitation learning / Pretraining / Supervised fine-tuning