OMP: One-step Meanflow Policy with Directional Alignment

Robot manipulation has increasingly adopted data-driven generative policy frameworks, yet the field faces a persistent trade-off: diffusion models suffer from high inference latency, while flow-based methods often require complex architectural constraints. Although in image generation domain, the MeanFlow paradigm offers a path to single-step inference, its direct application to robotics is impeded by critical theoretical pathologies, specifically spectral bias and gradient starvation in low-velocity regimes. To overcome these limitations, we propose the One-step MeanFlow Policy (OMP), a novel framework designed for high-fidelity, real-time manipulation. We introduce a lightweight directional alignment mechanism to explicitly synchronize predicted velocities with true mean velocities. Furthermore, we implement a Differential Derivation Equation (DDE) to approximate the Jacobian-Vector Product (JVP) operator, which decouples forward and backward passes to significantly reduce memory complexity. Extensive experiments on the Adroit and Meta-World benchmarks demonstrate that OMP outperforms state-of-the-art methods in success rate and trajectory accuracy, particularly in high-precision tasks, while retaining the efficiency of single-step generation.

We teach robots to physically manipulate objects using AI, but current systems face a stubborn trade-off: they are either too slow for real-world reactions or rely on overly complex designs. We wondered if the lightning-fast AI methods used to generate images could be adapted for robotics, but we found these direct applications often fail because they struggle to accurately learn physical speeds. To solve this, we developed a new framework called the One-step MeanFlow Policy (OMP). We took a fast generation model and introduced a mechanism that explicitly corrects the robot's predicted movements to align with the correct physical speed and direction. We also applied a mathematical shortcut to vastly reduce the computer memory needed to train the system. We tested our system on complex simulated robotic hands and arms and found it worked exceptionally well. It outperformed existing methods in both accuracy and success rates, proving that robots can achieve highly precise physical control without losing the ability to react instantly.