Paper 2602.07845
Recurrent-Depth VLA: Implicit Test-Time Compute Scaling of Vision-Language-Action Models via Latent Iterative Reasoning
- Published
- Feb 2026
- Research lab
- Independent
- Citations
- 11
- GitHub
- 42 stars
01 In brief
Summary
The paper introduces Recurrent-Depth VLA (RD-VLA), a Vision-Language-Action model that enables adaptive test-time compute via latent iterative reasoning.
Unlike token-based Chain-of-Thought methods, RD-VLA uses a weight-tied recurrent action head that refines a latent scratchpad over multiple iterations, keeping memory constant.
The architecture includes a Prelude, a Recurrent Core, and a Coda, trained with truncated backpropagation through time (TBPTT) and randomized recurrence depths.
At inference, an adaptive stopping criterion based on KL divergence between consecutive action predictions dynamically allocates compute.
Experiments on LIBERO show RD-VLA achieves 93.0% success with fixed iterations and 92.5% with adaptive computation, outperforming prior VLAs.
On CALVIN, it reaches an average task length of 3.39.
Real-world tests on a bimanual YAM robot show robust performance on tasks like towel folding and bread toasting.
The method provides up to 80× inference speedup over token-reasoning VLAs and scales log-linearly with compute depth, with task-dependent convergence profiles.
02 From the paper
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/