EfficientFlow: Efficient Equivariant Flow Policy Learning for Embodied AI

2025-12-01

AI Summary

Current visual-motor policy learning suffers from low data efficiency—requiring abundant expert demonstrations—and low sampling efficiency—yielding slow inference. To address these bottlenecks, we propose EfficientFlow, the first unified embodied policy learning framework that integrates equivariance into the flow matching paradigm. Theoretically, we prove that equivariant velocity networks preserve distributional equivariance, thereby enhancing data efficiency. We further replace the intractable acceleration term with a trainable proxy loss and introduce acceleration regularization to ensure stable, efficient training and rapid action sampling. EfficientFlow combines equivariant neural networks, isotropic Gaussian priors, and conditional trajectory modeling. Evaluated on diverse robotic manipulation tasks, it achieves state-of-the-art performance using only a few demonstrations, while significantly accelerating inference—demonstrating superior efficiency and strong generalization capability.

Technology Category

• Intelligent Robots: Learning & Optimization for ROB
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Generative modeling has recently shown remarkable promise for visuomotor policy learning, enabling flexible and expressive control across diverse embodied AI tasks. However, existing generative policies often struggle with data inefficiency, requiring large-scale demonstrations, and sampling inefficiency, incurring slow action generation during inference. We introduce EfficientFlow, a unified framework for efficient embodied AI with flow-based policy learning. To enhance data efficiency, we bring equivariance into flow matching. We theoretically prove that when using an isotropic Gaussian prior and an equivariant velocity prediction network, the resulting action distribution remains equivariant, leading to improved generalization and substantially reduced data demands. To accelerate sampling, we propose a novel acceleration regularization strategy. As direct computation of acceleration is intractable for marginal flow trajectories, we derive a novel surrogate loss that enables stable and scalable training using only conditional trajectories. Across a wide range of robotic manipulation benchmarks, the proposed algorithm achieves competitive or superior performance under limited data while offering dramatically faster inference. These results highlight EfficientFlow as a powerful and efficient paradigm for high-performance embodied AI.

Problem

• Improves data efficiency in generative visuomotor policies
• Accelerates action sampling for faster inference
• Enhances generalization with equivariant flow matching

Innovation

• Equivariant flow matching for improved data efficiency
• Acceleration regularization for faster action sampling
• Unified flow-based framework for efficient embodied AI