SeedPolicy: Horizon Scaling via Self-Evolving Diffusion Policy for Robot Manipulation

2026-03-05

AI Summary

Diffusion policies suffer significant performance degradation in long-horizon robotic imitation learning as the observation horizon increases. To address this limitation, this work proposes the Self-Evolving Gated Attention (SEGA) module, which recursively updates a time-varying latent state to compress long-horizon observations into a fixed-dimensional representation while filtering out redundant information. SEGA is compatible with both CNN and Transformer backbones and supports end-to-end training, effectively overcoming the temporal modeling bottleneck of diffusion policies with minimal parameter overhead. Evaluated on 50 tasks in RoboTwin 2.0, the method achieves an average performance gain of 36.8% in clean environments and a remarkable 169% improvement under challenging randomized conditions, matching the performance of billion-scale vision-language-action models with one to two orders of magnitude fewer parameters.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Intelligent Robots: Learning & Optimization for ROB
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Agentic search
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Imitation Learning (IL) enables robots to acquire manipulation skills from expert demonstrations. Diffusion Policy (DP) models multi-modal expert behaviors but suffers performance degradation as observation horizons increase, limiting long-horizon manipulation. We propose Self-Evolving Gated Attention (SEGA), a temporal module that maintains a time-evolving latent state via gated attention, enabling efficient recurrent updates that compress long-horizon observations into a fixed-size representation while filtering irrelevant temporal information. Integrating SEGA into DP yields Self-Evolving Diffusion Policy (SeedPolicy), which resolves the temporal modeling bottleneck and enables scalable horizon extension with moderate overhead. On the RoboTwin 2.0 benchmark with 50 manipulation tasks, SeedPolicy outperforms DP and other IL baselines. Averaged across both CNN and Transformer backbones, SeedPolicy achieves 36.8% relative improvement in clean settings and 169% relative improvement in randomized challenging settings over the DP. Compared to vision-language-action models such as RDT with 1.2B parameters, SeedPolicy achieves competitive performance with one to two orders of magnitude fewer parameters, demonstrating strong efficiency and scalability. These results establish SeedPolicy as a state-of-the-art imitation learning method for long-horizon robotic manipulation. Code is available at: https://github.com/Youqiang-Gui/SeedPolicy.

Problem

• Imitation Learning
• Diffusion Policy
• Long-horizon Manipulation
• Observation Horizon
• Robotic Manipulation

Innovation

• Self-Evolving Gated Attention
• Diffusion Policy
• Long-horizon Manipulation
• Imitation Learning
• Temporal Modeling