SE(3) Neural Potential Fields for 6-DoF Trajectory Planning Directly from Images Without Explicit 3D Reconstruction

📄 arXiv: 2609.24864v1 📥 PDF

作者: Jeffrey Eiyike, Masoud Ataei, Elvis Gyaase, Vikas Dhiman

分类: cs.RO, cs.AI

发布日期: 2026-09-21

💡 一句话要点

提出SE(3)神经势场以解决无显式3D重建的6自由度轨迹规划问题

🎯 匹配领域: 支柱三：空间感知与语义 (Perception & Semantics)

关键词: 神经网络 轨迹规划 6自由度 图像处理 机器人技术 势场学习 自动化

📋 核心要点

  1. 现有方法依赖于3D重建进行轨迹规划，导致计算复杂度高且精度不足。
  2. 提出的SE(3)神经势场直接从图像中学习，通过导航函数监督，避免了传统势场的缺陷。
  3. 在两个实验场景中，成功率达到90%，清晰度从不足1厘米提升至8.6-8.8厘米，碰撞率显著降低。

📝 摘要（中文）

在杂乱环境中达到6自由度抓取姿态需要无碰撞的轨迹，传统方法依赖于3D场景重建，导致精度和计算成本的增加。本文提出了一种从RGB图像中学习的SE(3)神经势场，通过导航函数进行监督，消除了传统方法的缺陷。实验结果表明，该方法在两个桌面场景中，轨迹规划的成功率高达90%，并且在执行路径时与真实几何体完全无碰撞，显著提高了清晰度和减少了臂链接接触率。

🔬 方法详解

问题定义：本文旨在解决在杂乱环境中实现6自由度抓取姿态时，传统方法依赖3D重建导致的高计算成本和精度不足的问题。

核心思路：提出了一种SE(3)神经势场，该势场直接从RGB图像中学习，并通过导航函数进行监督，从而消除了传统势场在吸引和排斥梯度相互抵消时的缺陷。

技术框架：整体架构包括图像输入、势场学习和轨迹规划三个主要模块。首先，利用RGB图像提取特征，然后通过神经网络构建势场，最后进行无碰撞轨迹规划。

关键创新：最重要的创新在于将神经网络与导航函数结合，形成了一种新的势场学习方式，避免了传统势场在复杂环境中的局限性。

关键设计：在网络结构上，采用了适合处理图像数据的卷积神经网络，并设计了特定的损失函数以优化势场的学习效果。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示，该方法在两个桌面场景中的执行抓取成功率为90.0%，而在仅依赖图像监督的情况下，成功率仅为40.0%。清晰度从不足1厘米提升至8.6-8.8厘米，臂链接触率从20.6-50.4%降至2.7-5.5%。规划时间显著缩短，从67-133秒降至约2秒。

🎯 应用场景

该研究的潜在应用领域包括机器人抓取、自动化仓储和服务机器人等。通过直接从图像中学习轨迹规划，能够显著提高机器人在复杂环境中的操作效率，降低对3D重建的依赖，具有广泛的实际价值和未来影响。

📄 摘要（原文）

Reaching a 6-DoF grasp pose in clutter requires a collision-free trajectory, conventionally obtained by reconstructing the scene in 3D and planning inside that reconstruction, at the cost of its accuracy and compute. Potential fields learned directly from images remove that dependency but inherit the classical weakness of artificial potential fields: where attractive and repulsive gradients cancel, the descent grazes the obstacle instead of going around it, and can stall short of the goal. We present an SE(3) neural potential field learned from posed RGB images and supervised with a navigation function, the geodesic distance to the grasp through free space recovered from those same images during training, which removes both failures. On two tabletop scenes, from obstacle-blocked starts executed on a UR10, the field converges within 3 cm of the grasp from every start and every path it executes is collision-free against the ground-truth geometry, against 25% and 0% under image supervision alone; mean clearance rises from under a centimeter to 8.6-8.8 cm and arm-link contacts fall from 20.6-50.4% to 2.7-5.5% of executed configurations. Executed grasp success is 90.0% and 40.0% on the two scenes, the residual failures being refusals of the Cartesian executor rather than of the field. Planning takes about 2 s against 67-133 s for RRT* on a reconstruction of the same images, though under a common offline harness the two are comparable: the deployed margin is the cost of collision-checking a dense reconstruction, not planner complexity.