NanoVLA: Routing Decoupled Vision-Language Understanding for Nano-sized Generalist Robotic Policies

Vision-language-action (VLA) models have significantly advanced robotic manipulation by integrating vision-language models (VLMs), and action decoders into a unified architecture. However, their deployment on resource-constrained edge devices, such as mobile robots or embedded systems (e.g., Jetson Orin Nano), remains challenging due to high computational demands, especially in real-world scenarios where power, latency, and computational resources are critical. To close this gap, we introduce Nano-scale Vision-Language Action (NanoVLA), a family of lightweight VLA architectures that achieve high performance with minimal resources. Our core innovations include: (1) vision-language decoupling that moves conventional early vision and language inputs fusion in VLM to late stage, achieving better performance while enabling caching and reduce inference overhead and latency; (2) long-short action chunking to ensure smooth, coherent multi-step planning without sacrificing real-time responsiveness; (3) dynamic routing that adaptively assigns lightweight or heavy backbones based on task complexity, further optimizing inference efficiency. Experimental results on several benchmarks, as well as real-world deployments, demonstrate that NanoVLA achieves up to 52x faster inference on edge devices compared to previous state-of-the-art VLA models, with 98% less parameters while maintaining or surpassing their task accuracy and generalization. Ablation studies confirm that our decoupling strategy preserves cross-task transferability, and the routing module enhances cost-performance trade-offs, enabling practical, high-precision robotic manipulation on resource-constrained hardware. The conventional approach in robot learning has been to train task-specific models from scratch, but vision-language-action (VLA) models are emerging as a transformative paradigm (Brohan et al., 2022; Zitkovich et al., 2023; Mittal et al., 2023; Bjorck et al., 2025; Cheang et al., 2025). This paradigm brings broad generalization to robot learning, but it clashes with the realities of deploying on resource-constrained edge hardware (e.g., Jetson Orin-class devices) for three key reasons: (1) inference remains slow and compute-intensive; Consequently, the existing VLA models remain impractical outside datacenter-class machines (Wang et al., 2025).Figure 1: Decoupled fusion for efficient VLA policies. This approach does enables better performance with less overhead and latency, which informs NanoVLA, small scale VLA that achieves better performance across both simulation and real-world tasks with only 2% of the parameter of models like OpenVLA, as shown in the Radar plot (right). In response, we introduce Nano-scale Vision-Language Action (NanoVLA), a framework that closes this deployment gap by reorganizing where modalities are fused, how actions are unrolled over time, and when a larger model backbone is invoked.