Abstract: In order to improve the computational efficiency and generalization performance of robot learning from demonstration， a robot learning from demonstration model was proposed based on generalization of dynamic movement primitives weighted by task parameters. The main steps were as follows：the dynamic movement primitives model was used to extract the characteristic parameters of the multi teaching motion trajectories； Under the new task parameter， the extracted feature parameterswas used to reconstruct the feature motion trajectory； the approximate degree of the teaching task parameters and the new task parameters were usedto weight the feature motion trajectory to generate a new motion track. Experiments on Kukaiiwa robot show that the proposed method may quickly and effectively generalize the new trajectory in the new task scenario. Compared with the existing methods， the proposed method has a great improvement in computational efficiency and generalization performance near the teaching task parameters. 

Key words: robot, learning from demonstration, dynamic movement primitives, generalization performance, task parameter

摘要： 为了提高机器人示教学习的计算效率以及泛化性能，提出了一种基于任务参数加权的动态运动基元泛化的机器人示教学习模型，主要步骤如下：运用动态运动基元模型提取多次示教运动轨迹的特征参数；在新的任务参数下，运用提取的特征参数重构特征运动轨迹；以示教任务参数与新任务参数的近似程度对特征运动轨迹进行加权叠加，生成新的运动轨迹。在Kukaiiwa机器人上的实验表明，在新的任务场景下，所提方法能够快速有效地泛化出新的运动轨迹，与已有方法相比，在计算效率及示教任务参数附近的泛化性能上有了较大的提升。

关键词: 机器人, 示教学习, 动态运动基元, 泛化性能, 任务参数