发表机构
School of Telecommunications Engineering, Xidian University; School of Computer Science and Technology, Xidian University; Tsinghua Shenzhen International Graduate School, Tsinghua University（西安电子科技大学通信工程学院; 西安电子科技大学计算机科学与技术学院; 清华大学深圳国际研究生院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究针对扩散策略实时部署计算需求大的问题，提出演进缓存调度（EVO）框架，通过进化搜索全局调度缓存刷新，引入冗余感知初始化和目标条件早期停止，大幅减少计算量并保留性能，实现动作生成加速和FLOPs降低。
AI 中文摘要
扩散策略通过迭代去噪动作块实现强大的视觉运动控制，但重复去噪使实时部署计算需求大。基于缓存的方法通过重用中间激活来降低推理成本，但现有无训练调度通常均匀分配计算，忽略块间异构冗余。我们引入演进缓存调度（EVO），通过进化搜索全局调度缓存刷新。它将候选者表示为块时间步晶格上的完整调度，可跳过冗余计算并保留闭环展开性能。还引入冗余感知初始化和目标条件早期停止。离线优化调度可直接插入预训练扩散策略。大量操纵基准表明，EVO在大幅减少计算的同时保留近全性能，动作生成加速高达8.05倍，FLOPs从15.77G降至1.96G。
英文摘要
Diffusion policies achieve strong visuomotor control by iteratively denoising action chunks, but repeated denoising makes real-time deployment computationally demanding. Cache-based methods reduce inference cost by reusing intermediate activations, but existing training-free schedules typically allocate computation uniformly across blocks, ignoring heterogeneous redundancy across blocks and leading to a suboptimal performance-efficiency trade-off. To bridge this gap, we introduce Evolving Cache Schedules (EVO), a training-free acceleration framework that globally schedules cache refreshes via evolutionary search. EVO represents each candidate as a complete schedule over the block-timestep lattice. Thus, redundant transformer computations during iterative denoising can be skipped through cache reuse while preserving closed-loop rollout performance. To make the search practical, EVO introduces redundancy-aware initialization, which seeds the population with promising schedules, and target-conditioned early stopping, which verifies and terminates once a desired performance target is reached. The offline-optimized schedule can be directly plugged into pretrained diffusion policies without retraining. Extensive manipulation benchmarks show that EVO preserves near-full performance while substantially reducing computation, achieving up to 8.05x action-generation speedup and reducing FLOPs from 15.77G to as low as 1.96G. Source code is available at https://github.com/pillom/EVO.
Comments15 pages, 3 figures, supplementary material included. Accepted by PRCV 2026
