Skip to content

论文arXiv 2611.174426 天前

预算条件VLN搜索:Bellman专家在部分可观测下优于TSP路径规划 ​

本文由 AI 根据公开内容整理,可能有误,请以原文为准:Racing the Clock: Teaching VLNs Budget-Conditioned Object Search(arXiv 2611.17442)

核心内容 ​

针对救援等时间敏感任务中视觉语言导航(VLN)模型缺乏时间预算适应能力的问题,研究提出预算条件目标搜索新范式。该任务要求智能体在显式动作预算内最大化目标实例发现数量。实验在HM3D数据集上进行,对比了拥有全图信息的特权TSP专家与基于高层跨场景语义先验的房间级Bellman专家对VLN模型的监督效果,旨在解决固定回合限制无法反映真实任务紧迫性的缺陷。

尽管TSP专家能规划理论最优路线,但其演示在部分可观测条件下对探索策略的学习指导价值较低。数据显示,在2000步(约500米)长预算测试中,Bellman监督的VLN模型目标发现率达71.4%,而TSP监督模型仅为65.5%,预训练基线为44.8%。这表明有效的探索监督应优化部分可观测下的决策期望回报,而非单纯模仿仅在环境完全已知时才成立的最优路径,且该优势随预算增加而扩大。

关键看点 ​

  • Bellman监督优势:在HM3D场景2000步预算下,Bellman监督VLN目标发现率达71.4%,比TSP监督高5.9个百分点,比预训练基线高26.6个百分点
  • 部分可观测适配:房间级Bellman专家利用跨场景语义先验最大化期望回报,解决了TSP专家因依赖全图真值导致在未知环境中探索策略泛化性差的问题
  • 预算条件机制:打破传统VLN固定回合评估限制,引入显式动作预算约束,使智能体能根据剩余时间动态调整从单点冲刺到广域探索的策略切换
  • 长时序性能增益:Bellman监督策略的有效性随动作预算增加而显著提升,验证了在长时间、大范围搜索任务中基于语义先验的探索优于全局路径模仿

工程与产业观察 ​

该研究修正了具身导航中“最优路径即最优探索”的认知偏差。对于机器人在未知环境的搜救或巡检任务,算法设计应从模仿全局规划转向优化部分可观测下的期望收益。工程落地时,建议采用融合语义先验的贝尔曼更新替代纯几何TSP作为教师信号,以提升机器人在时间约束和感知受限双重条件下的自主搜索效率。

英文摘要 (Abstract) ​

A rescue robot entering a building with thirty seconds on the clock should sprint toward the single most promising room; given ten minutes, it should instead explore a broader tour. Yet vision-language-navigation (VLN) models are typically evaluated under a fixed episode limit that the policy never observes, even though time-critical missions require behavior to adapt to the available budget. We study budget-conditioned object search: given a target category and an explicit action budget, an agent must find as many distinct target instances as possible before the budget expires. We investigate how to supervise VLNs for this task by comparing a privileged TSP expert, which plans near-optimal routes using the full map and true target locations, with a room-level Bellman expert that maximizes expected return under partial observability using high-level cross-scene semantic priors. Despite the TSP expert achieving substantially higher expert performance, its demonstrations are less effective for learning exploration: on held-out HM3D scenes, the Bellman-supervised VLN consistently outperforms the TSP-supervised VLN, with the advantage becoming more pronounced at longer budgets. At 2000 steps (500m), the Bellman-supervised VLN finds 71.4% of reachable target instances, compared with 65.5% for the TSP-supervised VLN and 44.8% for the pretrained VLN. These results highlight a distinction between optimal routing and optimal exploration: effective exploration supervision should optimize decisions under partial observability rather than imitate routes that are optimal only once the environment is fully known.

作者:Maitree Meher, Siming He, S. Shankar Sastry, Jitendra Malik, Claire Tomlin, Jonas Frey

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 2611.17442:Racing the Clock: Teaching VLNs Budget-Conditioned Object Search ↗访问 2026-11-16