# VARP视觉重排规划器实现98.3%成功率与0.4s推理耗时

> 原文链接：https://robotics.hellobike.com/papers/2026-11-04-2611-04093-3dc78a05
> 频道：papers | 更新日期：2026-11-04 | 标签：auto, arxiv
> 核心摘要：针对杂乱环境非单调物体重排难题，VARP框架通过视觉状态表征直接预测动作并在线自适应扰动。基准测试成功率达98.3%，单次规划仅0.4s，在严重干扰下仍保持94.5%成功率，有效规避传统搜索规划的高算力瓶颈。

# VARP视觉重排规划器实现98.3%成功率与0.4s推理耗时

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[VARP: A Visual Adaptive Rearrangement Planner for Non-Monotone Object Rearrangement Planning](https://arxiv.org/abs/2611.04093)（arXiv 2611.04093）

## 核心内容

针对杂乱环境中物体需临时移位才能到达目标位置的非单调重排任务，传统搜索规划器面临计算成本高及环境变化时需昂贵重规划的痛点。研究团队提出视觉自适应重排规划器VARP，直接从专家演示生成的视觉状态表征中学习策略。该方法在复杂重排基准测试中取得98.3%的任务成功率，平均规划耗时仅为0.4s，且在执行阶段遭遇持续严重干扰时仍能维持94.5%的成功率，验证了其在动态环境下的鲁棒性。

VARP架构基于当前与目标视觉状态表征直接预测抓取-放置动作，摒弃了在线组合搜索机制。系统引入动作细化策略（ARS）提升动作可行性，并结合进度强制策略（PES）缓解执行停滞问题。通过利用最新视觉表征持续更新动作指令，VARP实现了对环境扰动的在线自适应，无需从头重规划。实验表明该模型能有效泛化至未见过的物体数量场景，为具身智能操作提供了兼具实时性与抗干扰能力的端到端解决方案。

## 关键看点

- **核心性能**：在非单调重排基准测试中实现98.3%成功率，单任务规划耗时压缩至0.4s，显著优于传统搜索方法
- **抗扰机制**：集成动作细化策略(ARS)与进度强制策略(PES)，在执行期干扰及持续严重干扰下成功率分别达97.5%和94.5%
- **在线自适应**：基于视觉状态表征实时更新动作指令，无需在线组合搜索或全量重规划即可应对环境动态变化
- **泛化能力**：模型从专家演示视觉表征中学习，能够有效泛化至训练分布外未见过的物体数量场景

## 工程与产业观察

VARP证明了基于视觉表征的端到端学习在非单调长程操作任务中可替代高算力搜索规划。其0.4s级推理与强抗扰特性，为解决具身机器人在动态杂乱环境下的实时重排提供了新范式，对降低部署算力门槛及提升工业分拣、家庭整理等场景的工程落地性具有直接参考价值。

## 英文摘要 (Abstract)

Object rearrangement planning in cluttered environments is often non-monotone, requiring objects to be temporarily relocated before reaching their final destinations. Although search-based planners can solve such tasks, they incur substantial computational cost and require expensive replanning when the environment changes. We propose the Visual Adaptive Rearrangement Planner (VARP), which learns non-monotone rearrangement directly from visual state representations generated from expert demonstrations. VARP predicts pick-and-place actions from the current and goal visual state representations, and incorporates an Action Refinement Strategy (ARS) to improve action feasibility and a Progress Enforcement Strategy (PES) to mitigate progress stagnation. By continuously updating actions from the latest visual state representations, VARP adapts online to disturbances without replanning from scratch or performing online combinatorial search. In challenging rearrangement benchmarks, VARP achieves a 98.3\% success rate with only 0.4 s planning time per task, 97.5\% success under execution-time disturbances, and 94.5\% under continuous severe disturbances, while generalizing effectively to unseen object counts. These results demonstrate that VARP provides a scalable and robust solution for non-monotone object rearrangement.

作者：Hanwen Ren, Meng Zheng, Zhongpai Gao, Anwesa Choudhuri, Benjamin Planche, Terrence Chen, Ziyan Wu
:::

## 原始参考来源 (Sources)

- [arXiv 2611.04093：VARP: A Visual Adaptive Rearrangement Planner for Non-Monotone Object Rearrangement Planning](https://arxiv.org/abs/2611.04093)