IAAIRD-VLA
Recurrent-Depth VLA: Implicit Test-Time Compute Scaling of Vision-Language-Action Models via Latent Iterative Reasoning
Recurrent-Depth VLA gives vision-language-action models a weight-tied recurrent action head that refines a latent for any number of steps at constant memory, trained with truncated BPTT and stopped adaptively when the latent converges.
Inside the method
VLM features → [ Weight-tied action head ] × r (stop on convergence) → actions
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Adaptive depth
- Depth control
- Adaptive stopping based on latent convergence
- KV / state strategy
- Constant memory in recurrence depth
Reading note
Manipulation tasks that fail completely with one iteration exceed 90% success with four, while simple tasks saturate quickly. Reports up to 80× faster inference than prior reasoning-based VLAs.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{rdvla2026,
title = {Recurrent-Depth VLA: Implicit Test-Time Compute Scaling of Vision-Language-Action Models via Latent Iterative Reasoning},
author = {Yalcin Tur and Jalal Naghiyev and Haoquan Fang and Wei-Chuan Tsai and Jiafei Duan and Dieter Fox and Ranjay Krishna},
year = {2026},
eprint = {2602.07845},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2602.07845}
}