Vector Averaging-Based Fluctuation Suppression for 6DOF Estimation

针对DenseFusion模型直接输出值波动导致引导机械臂抓取任务失败问题，本文提出基于向量平均的6DOF估计值波动抑制方法。通过分析DenseFusion算法输出的6DOF估计结果，我们发现其波动符合正态分布，并提出了利用向量平均数对位置和姿态参数进行平滑处理的方法。实验结果表明，该方法能够有效抑制6DOF估计值的波动，显著提升了机械臂在不同姿态下的抓取成功率，从70%以上提升至90%以上。本文的研究为6DOF估计值的波动抑制提供了新的思路，并为手眼协同系统的实际应用提供了技术支持。To address the failure of robotic arm grasping tasks caused by fluctuations in the direct output values of the DenseFusion model, this paper proposes a vector averaging-based fluctuation suppression method for 6DOF estimation. By analyzing the 6DOF estimation results output by the DenseFusion algorithm, we observe that their fluctuations follow a normal distribution. Accordingly, we introduce a vector averaging approach to smooth both position and orientation parameters. Experimental results demonstrate that the proposed method effectively suppresses fluctuations in 6DOF estimation, significantly improving the robotic arm’s grasping success rate across different poses—from over 70% to above 90%. This study provides a novel approach for mitigating 6DOF estimation fluctuations and offers technical support for the practical application of hand-eye coordination systems.