Skip to content

论文arXiv 2611.190345 天前

SmolVLA策略推理分配研究:高频重规划优于动作块精细化 ​

本文由 AI 根据公开内容整理,可能有误,请以原文为准:Refinement vs. Replanning: Inference Compute Allocation in Robot Policies(arXiv 2611.19034)

核心内容 ​

本文探讨了冻结权重的分块机器人策略(Chunked Robot Policy)在不更新模型参数的前提下,如何优化部署阶段的推理算力分配。研究核心在于权衡两种耦合机制:增加单次策略调用的计算量以精细化预测动作块(Refinement),或提高策略调用频率以基于新观测进行重规划(Replanning)。评估基于SmolVLA检查点在同步LIBERO仿真环境中的表现,采用去噪计数作为精细化控制变量,并固定动作块长度K=50,通过标称计算速率与完整尝试成本双重指标量化性能。

实验数据显示算力成本构成具有显著差异:在去噪步数d=10时,82.1%的算术运算独立于去噪过程,而主机时间仿射拟合显示仅17.7%为固定开销。在h=2配置下,d=2的成功率(139/160)比d=5高出10.6个百分点,且每控制步算术量减少5.9%。关键发现是当去噪步数固定为d=2时,将调用频率提升至h=5虽使成功率微降至138/160,但单次尝试的算术运算量大幅下降60.7%。这证明对于特定任务,提高反馈频率比单纯增加生成努力更具性价比,部署适配需联合评估两者而非寻求通用设置。

关键看点 ​

  • 算力成本解耦:d=10时82.1%算术运算与去噪无关,主机时间固定开销占比17.7%,揭示非生成计算主导推理延迟
  • 精细化边际递减:h=2时d=2成功率139/160优于d=5的122/160,且单步算术量降低5.9%,过度去噪反损性能
  • 高频重规划优势:d=2时将调用频率从h=2提至h=5,成功率仅降0.6%但单次尝试算术量骤降60.7%
  • 评估方法论:提出结合标称计算速率与完整尝试成本的联合评估流程,避免单一指标误导部署决策

工程与产业观察 ​

该研究为具身智能端侧部署提供了关键工程指引:在算力受限场景下,与其追求单次推理的极致精度(如增加扩散模型去噪步数),不如维持较低生成质量并大幅提升感知-决策闭环频率。这种“粗生成+快反馈”策略在保持任务成功率的同时显著降低总计算负载,对VLA类大模型在嵌入式平台的实时化落地具有直接参考价值,强调了系统级推理预算分配的重要性。

英文摘要 (Abstract) ​

A frozen chunked robot policy can spend deployment-time inference compute in two coupled ways: more work within each policy call to refine a predicted action chunk, or more frequent policy calls to replan from fresh observations. We study this allocation without updating model weights. Our evaluation procedure reports success with both nominal compute rate and complete attempt cost, selects comparisons on discovery tasks, and tests prespecified comparisons on held-out tasks. Comparisons added after inspecting the held-out results only summarize observed patterns. We demonstrate the procedure with one frozen SmolVLA checkpoint in synchronous LIBERO simulation, using denoising count as the refinement control and fixing chunk length at K=50. Arithmetic and latency characterize cost differently: at d=10, 82.1% of counted arithmetic is independent of denoising, whereas a descriptive affine fit to fixed-input host time assigns 17.7% to its intercept. At h=2, d=2 succeeds on 139/160 attempts versus 122/160 for d=5 (+10.6 percentage points; task-conditional 95% interval: +4.4 to +16.9), while using 5.9% less arithmetic per control step. Executing h=5 rather than h=2 at d=2 yields 138/160 versus 139/160 successes and 60.7% lower arithmetic per attempt; equivalence is not established. The case study shows why deployment adaptation must evaluate generation effort and feedback frequency jointly; it does not define a universal setting.

作者:Alba Maria Tellez Fernandez

基于本页技术方向关联的整机参数、核心零部件、实测报告与同赛道选型横评。

外部来源

本页事实对应的公开页面,以原文为准。(点击将离开本站)

  1. arXiv 2611.19034:Refinement vs. Replanning: Inference Compute Allocation in Robot Policies ↗访问 2026-11-18