Dynamic Movement Primitives (DMP) is a trajectory encoding method based on imitation learning, which effectively simulates and generalizes human-demonstrated trajectories. However, it is essential to equip the DMP system with obstacle avoidance capabilities to allow the robotic manipulator to apply trajectories learned from obstacle-free scenarios to environments with obstacles, including multiple obstacles, for task execution. In this work, we first denoised and smoothed the collected demonstration trajectories to ensure the optimal motion characteristics of the input data. We then proposed an enhanced DMP-based obstacle avoidance scheme that integrates superquadratic and Gaussian potential functions to create a composite potential function, combining the shape adaptability and smoothness of both potential functions. To improve adaptability to differently shaped obstacles, we also developed an adaptive weighting coefficient based on the distance between the robotic manipulator’s end-effector and obstacles. By computing the dynamic potential function and integrating its negative gradient into our DMP model, we achieve dynamic adjustments to the changing environmental conditions of the robotic manipulator’s motion space, enhancing adaptability and accuracy in navigating around obstacles. Experimental results demonstrate that our method effectively avoids obstacles smoothly without abrupt trajectory changes.