Search papers, labs, and topics across Lattice.
This paper introduces the infer-diagnose-refine (IDR) framework, which enhances vision-language-action (VLA) models by dynamically adjusting the importance of visual observations during test-time action prediction. By inferring actions under both factual and counterfactual scenarios, the framework diagnoses causal effects to refine predictions without requiring additional training. Experimental results reveal significant performance improvements across various VLA architectures, highlighting the framework's effectiveness in addressing the challenges of modality fusion in robot manipulation tasks.
Dynamically adjusting visual importance at test time can significantly enhance action prediction accuracy in VLA models.
Vision-language-action (VLA) models predict sequential actions to execute tasks specified by language instructions, conditioned on visual observations and proprioceptive states. However, how to fuse modalities in VLA models remains an open problem, since robot manipulation involves dynamic phases, such as long-distance movements and close-range interactions, in which the importance of visual observations may vary over time. In this paper, we propose an infer-diagnose-refine (IDR) framework, a model-agnostic framework that can be integrated with diverse VLA architectures for refining action predictions at test time. IDR first infers actions under factual and counterfactual scenarios of visual observations, and then diagnoses the causal effects of visual observations as the estimated dynamic importance, which is finally used to refine the action predictions in a training-free manner. We further design a causality-aware action refiner to realize the IDR framework, including zero-padding interventions for inferring counterfactual actions, norm-based quantification for diagnosing causal effects, and gated residual fusion for refining actions. Extensive experiments on both simulation benchmarks and real-world tasks show improvements in overall performance across multiple VLA backbones, demonstrating the efficacy of dynamically adjusting visual importance at test time.
