In context
In 2024, robotic arm skill learning and generalization were active research areas, driven by the need for higher autonomy in complex industrial tasks. While obstacle avoidance algorithms were widely used, they often distorted the learned trajectory shape, reducing similarity to the demonstrated path—a critical issue for precision applications.
What was reported
Researchers from Xiamen University proposed a trajectory learning and obstacle avoidance method based on segmented dynamic movement primitives (DMP). The method first encodes a demonstration trajectory using DMP to generate a learned trajectory. Then, the rapidly-exploring random tree (RRT) algorithm is used to find an obstacle-free path in the workspace that smoothly returns to the original trajectory.
Key to the approach is selecting intermediate points from the RRT path to segment the DMP encoding. This segmented learning generates a playback trajectory that retains the original trajectory's shape features while avoiding obstacles. The method was validated through handwritten letter experiments and object handling tests on a six-axis robotic arm platform.
Results showed that the segmented DMP encoding effectively solved trajectory deformation and feature destruction caused by traditional obstacle avoidance algorithms. The reproduced trajectories had significantly higher similarity to the demonstration trajectories compared to conventional methods.
Why it mattered
This work addressed a practical limitation in industrial automation: maintaining trajectory fidelity during obstacle avoidance. By combining DMP with RRT and segmentation, it offered a way to improve the precision of learned motions in dynamic environments, potentially enhancing the reliability of robotic arms in tasks like assembly or material handling where path consistency is critical.
“The experimental result trajectory shows a significant improvement in similarity to the demonstration trajectory compared to the traditional obstacle avoidance algorithms.”
Source: 《机器人》期刊 (robot.sia.cn) · Published 2024-05-08 · “基于分段动态运动基元的机械臂轨迹学习与避障方法”