Robion is a new serving and management system designed to run Vision‑Language‑Action (VLA) models on multi‑GPU edge servers for robot factories. It splits the VLM and ADiT stages within a single GPU, shares streams across multiple models, and prioritizes requests by remaining SLO time, enabling high robot load while meeting strict latency requirements. In experiments, Robion achieves 6.7× higher robot load than vLLM‑Omni and 1.5× higher than a monolithic pipeline, and can serve 64 robots on a 4‑GPU server with 98% SLO attainment.
arXiv:2606. 08094v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) policies are typically shipped as Python/PyTorch stacks that assume a workstation-class GPU, a mismatch for the hardware on which robots actually run.
arXiv:2608. 14586v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models are becoming a promising paradigm for autonomous driving, but their deployment on existing vehicle platforms remains difficult because they introduce both high inference latency and strong GPU-side resource pressure.
arXiv:2606. 07383v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models have shown strong potential for robotic manipulation, but real-time deployment on edge hardware remains challenging.
arXiv:2607. 16074v1 Announce Type: cross Abstract: The post-training of Vision-Language-Action (VLA) models is essential due to the diversity of simulators, robot embodiments, and task objectives.
arXiv:2610.34717v1 Announce Type: cross Abstract: Vision--language--action (VLA) models increasingly run on edge devices aboard robots, where inference latency constrains closed-loop robot control. H...
arXiv:2608. 03682v1 Announce Type: new Abstract: Physical AI policies require inference throughout their lifecycle, including model evaluation, cloud reinforcement learning rollout, edge GPU serving, and onboard deployment.
OxyGen introduces a unified KV cache management system for Vision‑Language‑Action (VLA) models that run multiple tasks in parallel, such as manipulation, conversation, and memory construction. By treating the KV cache as a shared resource, it eliminates redundant prefill operations and enables cross‑frame continuous batching, leading to significant speedups—up to 3.7× faster than isolated execution—while maintaining high language throughput and action frequency on both RTX 4090 and Jetson AGX Thor platforms. The approach is validated across several backbones and expert configurations, including deployment on a humanoid robot with on‑board Jetson hardware.
arXiv:2512. 22560v2 Announce Type: replace-cross Abstract: Agentic Reinforcement Learning (RL) trains LLMs through multi-turn interactions with environments, producing workloads that mix compute-bound prefill, bandwidth-bound decoding, CPU-heavy environment execution, and bursty reward evaluation.
The paper introduces CloudEdgeVLA, a cloud‑edge policy for Vision‑Language‑Action models that treats temporal misalignment as a representation‑learning problem. It encodes delayed observations into slowly varying task features on the cloud while a lightweight edge head fuses the latest cloud feature with current local vision. Experiments on four LIBERO suites show that CloudEdgeVLA retains 63.8–78.0% success under a 40‑step delay window, far outperforming VLASH and single‑path baselines.
Visual Perception Engine (VPEngine) is a modular framework that enables efficient GPU usage for robotic vision tasks by sharing a foundation model backbone across multiple specialized task heads. It eliminates redundant feature extraction, supports dynamic task prioritization, and achieves up to 3× speedup over sequential execution. The open‑source Python implementation, with ROS2 C++ bindings, delivers real‑time performance (≥50 Hz) on NVIDIA Jetson Orin AGX using TensorRT‑optimized models.
The paper introduces a framework for Flow‑Matching Vision‑Language‑Action (VLA) models that allows independent adjustment of backbone depth, action expert depth, and denoising steps. Lightweight Exit Transformers are added at intermediate layers to enable early exits, and a KV Cache synthesis mechanism manages skipped layers so the action expert can exit deeper than the backbone. Experiments on SmolVLA and π0.5 across LIBERO and Meta‑World show that joint tuning of these compute axes reduces latency by 79.2 % and FLOPs by 31.8 %, while improving mean success rate by 5.6 %.
EAServe introduces an encode-aware disaggregated serving framework for multimodal large language models (MLLMs), restructuring the traditional Prefill-Decode pipeline into a three-stage Encode-Prefill-Decode (EPD) system. By treating Encode as the control point, EAServe coordinates load‑adaptive micro‑batching, rate‑controlled offloading to prefill workers, and dynamic SM partitioning to balance GPU utilization across stages. Its Hybrid Auto Selection (HAS) layer optimizes GPU allocation, encode batch size, and offload ratio using capacity profiling and Bayesian optimization, achieving up to 4.3× higher goodput compared to NVIDIA Dynamo and 1.7× higher than vLLM on various MLLM architectures.