Abstract: Aiming at the problem of low similarity between the planned robotic arm movement trajectory and the demonstration trajectory in complex work scenes with obstacles, a trajectory learning and obstacle avoidance method based on segmented DMP (dynamic motion primitive) is proposed. Firstly, the DMP model is adopted to encode the demonstration trajectory to generate a learning trajectory, and the rapidly-exploring random tree (RRT) is utilized also to obtain an obstacle avoidance path in the workspace that can return to the original trajectory smoothly. Then, the intermediate point of obstacle avoidance path is determined by segmented DMP optimization coding to learn and generate the playback trajectory with generalization ability. Finally, the robotic arm reproduces the trajectory, achieving obstacle avoidance while preserving the original trajectory characteristics. The handwritten letters experiment and the object handling experiment on the six-axis robotic arm platform show that the trajectory deformation and feature destruction caused by the traditional obstacle avoidance algorithms, is effectively solved by DMP segmentation coding. The experimental result trajectory shows a significant improvement in similarity to the demonstration trajectory compared to the traditional obstacle avoidance algorithms, which verifies the effectiveness of the method proposed.