SpikingVLA: Asynchronous Spiking Vision-Language-Action Models

Jingya Wang; Dehao Zhang; Shuai Wang; Malu Zhang; Yang Yang; Haizhou Li

cs.CLOct 7, 2026

SpikingVLA: Asynchronous Spiking Vision-Language-Action Models

Authors: Jingya Wang, Dehao Zhang, Shuai Wang, Malu Zhang, Yang Yang, Haizhou Li

Organizations: University of Electronic Science and Technology of China · Shenzhen Loop Area Institute · The Chinese University of Hong Kong (Shenzhen)

Abstract

ANN-to-SNN conversion offers a practical route toward energy-efficient spiking Vision-Language-Action (VLA) models by bypassing the substantial cost of training large-scale SNNs from scratch. However, existing methods often require many timesteps to maintain competitive performance, resulting in substantial inference latency for real-time VLA deployment. To address this challenge, we introduce SpikingVLA, an ANN-to-SNN conversion framework that enables accurate and low-latency spiking VLA inference. Specifically, we propose a Dendritic Integrate-and-Fire (DIF) neuron that alleviates channel-wise activation outliers through dendritic mixing and adaptive somatic firing, enabling accurate ANN-to-SNN conversion with fewer timesteps. Building on DIF neurons, we further introduce an asynchronous execution mechanism that overlaps temporal computation across VLA components, reducing synchronization overhead and latency. Extensive experiments demonstrate that SpikingVLA achieves competitive navigation performance with substantially improved inference efficiency. Compared with existing spiking VLA methods, SpikingVLA improves SR and SPL by 11.9% and 12.6%, respectively, while reducing first-action latency by 11.2 $\times$ . These results establish SpikingVLA as a practical framework for deploying pretrained VLA models with high-performance and low-latency spiking inference.

Figures & tables

Figure 1: (a) Pronounced activation outliers across VLA components increase spike encoding errors at limited timesteps, degrading conversion accuracy. (b) Non-uniform temporal horizons across components introduce synchronization stalls and idle periods, increasing inference latency.

Figure 2: Channel-wise activation imbalance across VLA components.

Figure 3: (a) Computation flow of the VLA model. (b) DIF neuron with dendritic feature integration and channel-adaptive spike encoding. (c) Differential conversion strategies for different operators. (d) Event-driven inference by accumulating differential responses over timesteps.

Figure 4: Performance and Efficiency analysis. (a) Navigation trajectories and action-prediction convergence on the VLN-CE R2R and RxR Val-Unseen splits. (b) Inference-latency comparison, highlighting the reduction achieved by asynchronous spike-based execution.

Type	Method	TF	R2R Val-Unseen				RxR Val-Unseen				Total T $\downarrow$
			NE $\downarrow$	OS $\uparrow$	SR $\uparrow$	SPL $\uparrow$	NE $\downarrow$	SR $\uparrow$	SPL $\uparrow$	nDTW $\uparrow$
ANN	StreamVLN ( Wei et al., 2025 )	✗	4.98	64.2	56.9	51.9	6.22	52.9	46.0	61.9	-
	InternVLA-N1 ( Wei et al., 2026 )	✗	4.05	70.7	64.3	58.5	4.58	61.4	51.8	70.0	-
	ETPNav ( An et al., 2024 )	✗	4.71	65.0	57.0	49.0	5.64	54.7	44.8	61.9	-
	HNR ( Wang et al., 2024 )	✗	4.42	67.0	61.0	51.0	5.50	56.3	46.7	63.5	-
	NaVILA ( Cheng et al., 2024 )	✗	5.28	61.5	53.9	49.3	6.12	52.3	46.1	61.0	-

Table 1: Navigation performance on the R2R-CE Val-Unseen and RxR-CE Val-Unseen. TF indicates training-free paradigms, and Total T denotes the total number of full-network timesteps. Bold values indicate the best results among SNN-based methods.

Type	Method	TF	Closed-Loop Navigation				Deployment Efficiency
Type			NE $\downarrow$	OS $\uparrow$	SR $\uparrow$	SPL $\uparrow$	Mem. $\downarrow$	TTFA $\downarrow$	Total T $\downarrow$
ANN	NaVILA-Blind ( Cheng et al., 2024 )	✗	6.03	49.0	36.2	33.3	16126.2	-	-
	NaVILA-Vision ( Cheng et al., 2024 )	✗	5.49	58.7	50.2	45.5	16126.2	-	-
	NaVILA-R ( Song et al., 2026 )	✗	6.29	52.1	36.5	29.5	16126.2	-	-
	NaVILA-AWQ ( Cheng et al., 2024 )	✗	6.58	48.2	32.8	27.4	5846.9	-	-
SNN	SpikeVLA ( Song et al., 2026 )	✗	6.02	53.6	32.7	28.5	6251.5	186.1	500-600

Table 2: Closed-loop navigation performance and deployment efficiency on VLN-CE-Isaac.

Figure 5: (a) Channel-wise activation distributions. (b) Layer-wise conversion discrepancy under different simulation timesteps.

Module	Neuron	Number of Thresholds
Module	Neuron	$\mathbf{n=1}$	$\mathbf{n=2}$	$\mathbf{n=4}$	$\mathbf{n=8}$
Vision Encoder	MT	1.13	0.99	0.74	0.52
Vision Encoder	DIF	1.15	0.74	0.39	0.11
LLaMA Decoder	MT	0.48	0.40	0.31	0.24
LLaMA Decoder	DIF	0.52	0.36	0.22	0.11

Table 3: Module-wise ANN–SNN conversion error under different spiking neuron configurations.

Appendix figures & tables6 assets

Supplementary material from the paper’s appendix.

Appendix

Method	Vision Encoder	Projector	Language Decoder	Actor Network	Total
NaVILA-Vision (ANN)	24.53	0.51	120.33	2.90	148.27
SpikingVLA ( $T=4$ )	3.87	0.31	19.05	0.45	23.68
SpikingVLA ( $T=8$ )	7.70	0.35	37.84	0.91	46.80

Appendix

Table 4: Module-wise theoretical energy consumption breakdown of VLA inference. Energy is estimated using the MAC/AC-based model under a 45-nm CMOS technology.

	Vision Encoder			LLaMA Decoder
$T$	Uncalib.	MT-Calib.	DIF-Calib.	Uncalib.	MT-Calib.	DIF-Calib.
1	1.00	1.13	1.15	1.30	0.48	0.52
2	1.02	0.99	0.74	1.31	0.40	0.36
4	1.07	0.74	0.39	1.26	0.31	0.22
8	1.12	0.52	0.19	1.24	0.24	0.11

Appendix

Table 5: Effect of calibration on ANN–SNN representation fidelity. Module-wise representation discrepancy is reported under different simulation timesteps

T

. Lower is better.

Figure 6: Activation redistribution across LLaMA decoder layers.

Figure 7: Component-wise differential conversion of the vision encoder and LLaMA decoder. (a) Vision encoder with DIF-based conversion of attention and feed-forward operations. (b) LLaMA decoder with RoPE-based attention and SwiGLU. Affine, nonlinear, and multiplicative operators are handled by their corresponding differential formulations.

Figure 8: Representative navigation rollouts of SpikingVLA.

Figure 9: Temporal convergence of action predictions toward the corresponding ANN outputs across internal timesteps, with different inputs exhibiting distinct convergence rates.

Explore similar work

SpikeVLA: Vision-Language-Action Models with Spiking Neural Networks
Jun 26, 2026Ruiqi Song, Dujun Nie, Siyu Teng +7Diffusion-Based Vision-Language-Actions Fast Inference
College of Surveying and Geo-informatics, Tongji University, Shanghai, China · State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China · OpenSpace Lab, China +6
VL2Spike: Spike-driven Distillation from VLMs for Low-Power Visual Perception in Embodied AI
Jun 14, 2026Zinan Liu, Eric Zheng, Soumyaratna Debnath +3Spiking Neural Networks Time-To-First-Spike
School of EEE, Nanyang Technological University (NTU), Singapore · Department of Computer Science, University of Toronto and Advanced Micro Devices, Inc., Canada · State Key Laboratory of Extreme Photonics and Instrumentation, Zhejiang University, Hangzhou 310027, China +1
Understanding Asynchronous Inference Methods for Vision-Language-Action Models
May 4, 2026Ayoub AgouzoulInference Latency Fast Inference
École polytechnique · Systems Group, ETH Zürich

Jun 26, 2026, cs.ROJ/K move · Enter open · S save

SpikeVLA: Vision-Language-Action Models with Spiking Neural Networks

Ruiqi Song, Dujun Nie, Siyu Teng, Baiyong Ding+6

College of Surveying and Geo-informatics, Tongji University, Shanghai, China · State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China · OpenSpace Lab, China+6

Diffusion-Based Vision-Language-Actions Fast Inference Time-To-First-Spike Embodied Artificial Intelligence Vision Encoders Neural Network

Open paper AskPDF arXiv

Vision-Language-Action (VLA) models have become a dominant paradigm for embodied intelligence. However, most existing approaches are built on large-scale transformers, resulting in substantial inference latency and energy consumption that limit their practical deployment in low-power, real-time scenarios. We propose SpikeVLA, a spiking VLA architecture for embodied navigation with energy-efficient inference, consisting of three key components. (i) A spiking vision encoder, Spike-V, that replaces dense continuous layers with event-driven spiking layers to reduce the energy consumption of visual representation learning. (ii) A multi-modal spiking large language model, Spike-L, that reformulates cross-modal reasoning with spiking dynamics and token-level event-driven sparsity to further lower computational cost. (iii) A spiking action policy network, Spike-A employs Laplacian-kernel population coding with a multi-layer fully connected SNN, and decodes spiking activities into stable and robust continuous control with energy-efficient inference under low-power constraints. Experiments on navigation and robotic control tasks show that SpikeVLA significantly reduces energy consumption and computational cost while maintaining competitive performance, highlighting its potential for low-power, real-time embodied intelligence.

Abstract

Explore similar work

SpikeVLA: Vision-Language-Action Models with Spiking Neural Networks

VL2Spike: Spike-driven Distillation from VLMs for Low-Power Visual Perception in Embodied AI

Understanding Asynchronous Inference Methods for Vision-Language-Action Models