In context
As human-robot collaboration expanded into shared workspaces, the need for robots to behave in predictable, human-like ways became critical for safety and efficiency. Traditional control methods often produced non-intuitive motions that hindered interaction, especially for non-expert users.
What was reported
Researchers proposed a human-in-the-loop learning framework to teach redundant manipulators human-like posture generation. Using an exoskeleton master device, human arm movements were captured and mapped to a 7-DOF robot via a CLIK-based controller with a null-space posture constraint. An encoder-decoder neural network learned the mapping from state to posture control actions, with a self-validation mechanism to detect errors.
To address covariate shift, an online relabeling method was introduced, reducing total demonstration time to under 10 minutes. Experiments compared human-robot postures and tracked dynamic trajectories, validating the approach. User studies showed that manipulators with human-like posture constraints were perceived as more friendly and acceptable as collaborative tools for non-professionals.
Why it mattered
This framework offers a practical path to making industrial manipulators more intuitive and safer in human-centric tasks, potentially easing adoption in SMEs and collaborative assembly lines where non-expert operators are common.
“Endowing robotic manipulators with human-like actions can make their behaviours more explainable and predictable, which improves the quality and safety of human-robot collaboration.”
Source: 《机器人》期刊 (robot.sia.cn) · Published 2023-09-16 · “一种用于机械臂拟人化控制的学习框架”