aiqbench analysis
HybridFlow advances real-time robotic manipulation by optimizing generative policies to achieve high accuracy with significantly reduced inference latency on edge hardware like the Jetson AGX Thor. The approach balances action precision and speed through a novel three-stage procedure that reuses a single model, outperforming standard diffusion policies in both success rates and task performance metrics. This work is critical for deploying autonomous robots that require rapid decision-making without relying on heavy cloud dependencies.
HybridFlow: A 2-NFE Generative Policy for Real-Time Robotic Manipulation
arXiv:2602.13718v2 Announce Type: replace-cross Abstract: Generative policies for robotic manipulation must balance action accuracy with inference latency. We present HybridFlow, a three-stage policy inference procedure requiring two network function evaluations (2-NFE). A Global Jump uses the MeanFlow average velocity to generate a coarse action trajectory; a parameter-free ReNoise interpolation constructs a state at a nonzero refinement time; and a Local Refine queries the instantaneous-velocity limit of the same network at that time. This construction reuses a unified model without distillation. Our analysis characterizes interval-composition errors and the attenuation of endpoint error under ReNoise interpolation. Controlled RoboMimic ablations support the MeanFlow proposal and intermediate-state construction, achieving 95% average success with reused noise and 95.5% with fresh noise, versus 78% for one-step MeanFlow. Across five real-robot settings with all policies running on the same Jetson AGX Thor, HybridFlow improves normalized task performance over 16-step Diffusion Policy by 13-68 points with approximately eightfold lower action-generation latency. Additional experiments demonstrate its compatibility as an action expert in a vision-language-action framework. Project page: https://hybridflow-anonymous.pages.dev/
Abstract:Generative policies for robotic manipulation must balance action accuracy with inference latency. We present HybridFlow, a three-stage policy inference procedure requiring two network function evaluations (2-NFE). A Global Jump uses the MeanFlow average velocity to generate a coarse action trajectory; a parameter-free ReNoise interpolation constructs a state at a nonzero refinement time; and a Local Refine queries the instantaneous-velocity limit of the same network at that time. This construction reuses a unified model without distillation. Our analysis characterizes interval-composition errors and the attenuation of endpoint error under ReNoise interpolation. Controlled RoboMimic ablations support the MeanFlow proposal and intermediate-state construction, achieving 95% average success with reused noise and 95.5% with fresh noise, versus 78% for one-step MeanFlow. Across five real-robot settings with all policies running on the same Jetson AGX Thor, HybridFlow improves normalized task performance over 16-step Diffusion Policy by 13-68 points with approximately eightfold lower action-generation latency. Additional experiments demonstrate its compatibility as an action expert in a vision-language-action framework. Project page: this https URL
9 pages. Updated author list and title; expanded analysis, controlled ablations, real-robot evaluation, and VLA experiments. Project page: this https URL
Robotics (cs.RO); Artificial Intelligence (cs.AI)
arXiv:2602.13718 [cs.RO]
(or arXiv:2602.13718v2 [cs.RO] for this version)
Full text under the publisher's open licence · 200 words
Related coverage
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026
NVIDIA GTC Taipei at COMPUTEX: Live Updates on What’s Next in AI
NVIDIA · May 21, 2026
The Machines that Make the Machines
NVIDIA · Oct 7, 2026
Build a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod
AWS · Sep 4, 2026