Recurrent-Depth VLA: Implicit Test-Time Compute Scaling of Vision-Language-Action Models via Latent Iterative Reasoning

TL;DR

RD-VLA introduces a recurrent architecture for vision-language-action models that adaptively scales compute at test time through latent iterative refinement, enabling efficient and scalable reasoning in robotics tasks.

Contribution

It proposes a novel recurrent, weight-tied action head that allows variable inference depth with constant memory, replacing token-based reasoning with latent iterative refinement.

Findings

Recurrent depth improves success rates from 0% to over 90% in complex tasks.

Constant memory usage enables up to 80x faster inference.

Adaptive stopping criterion effectively allocates compute based on task complexity.

Abstract

Current Vision-Language-Action (VLA) models rely on fixed computational depth, expending the same amount of compute on simple adjustments and complex multi-step manipulation. While Chain-of-Thought (CoT) prompting enables variable computation, it scales memory linearly and is ill-suited for continuous action spaces. We introduce Recurrent-Depth VLA (RD-VLA), an architecture that achieves computational adaptivity via latent iterative refinement rather than explicit token generation. RD-VLA employs a recurrent, weight-tied action head that supports arbitrary inference depth with a constant memory footprint. The model is trained using truncated backpropagation through time (TBPTT) to efficiently supervise the refinement process. At inference, RD-VLA dynamically allocates compute using an adaptive stopping criterion based on latent convergence. Experiments on challenging manipulation tasks show that recurrent depth is critical: tasks that fail entirely (0 percent success) with single-iteration inference exceed 90 percent success with four iterations, while simpler tasks saturate rapidly. RD-VLA provides a scalable path to test-time compute in robotics, replacing token-based reasoning with latent reasoning to achieve constant memory usage and up to 80x inference speedup over prior reasoning-based VLA models. Project page: https://rd-vla.github.io/