基于GMM与JA的轨迹学习与泛化方法研究

Learning and Generalization of Trajectories Based on GMM and JA

摘要: 本文针对机器人拆解任务中路径起点或终点发生变化时重定位时间成本高、编程效率低的问题，提出一种基于高斯混合模型(GMM)与Jerk Accuracy模型(JA)的轨迹学习与泛化方法。首先，通过高斯混合模型和高斯混合回归获得最优示教轨迹，然后引入JA模型，从优化角度生成具有泛化能力的复现轨迹，实现任务位置约束下起点或终点轨迹的泛化。最后，设计仿真实验对所提出方法进行验证。结果表明：该方法有效解决了上述问题，相较于传统的GMM-DMP方法，实验结果显示泛化轨迹与示教轨迹的相似性有了明显提高，验证了所提方法的有效性。

Abstract: To address the issues of high relocation time costs and low programming efficiency in scenarios where the start or end points of paths change in robotic disassembly tasks, this study proposes a trajectory learning and generalization method based on Gaussian Mixture Model (GMM) and Jerk Accuracy (JA) model. Firstly, the optimal demonstration trajectory is obtained through the Gaussian Mixture Model and Gaussian Mixture Regression. The JA model is then introduced to optimize and generate reproduction trajectories with generalization capabilities, allowing for trajectory generalization under task position constraints at the start or end points. Finally, a simulation is designed to validate the proposed method. The results demonstrate that this method effectively solves the aforementioned issues, significantly improving the similarity between the experiment trajectories and the demonstration trajectories compared to traditional GMM-DMP trajectories, thus verifying the method’s effectiveness.