Hierarchical Learning for Robotic Assembly Tasks Leveraging Learning from Demonstration

Robotic assembly in manufacturing settings is a special type of long-horizon task and motion planning problem. While devising a motion plan for the robot is itself challenging, identifying a task and learning it adds to problem's complexity. This article proposes a hierarchical learning -based approach that leverages the multilevel structure to seamlessly integrate task identification and sequencing with robot motion planning. Given the final assembly goal the higher-level agent emphasizes comprehending tasks and learning task plans. It generates sequences of sub-tasks, while the lower-level agent concentrates on executing the current sub-task. The higher-level agent employs a goal driven reinforcement learning approach to learn the sequencing task, allowing it to adapt to unseen assemblies. Meanwhile, the lower level adopts a learning from demonstration approach for motion planning, which can learn primitive skills from one-time demonstration and intelligently combine these primitive skills for complicated tasks. The critical contribution of this work lies in the development of a novel method capable of comprehending and executing long horizon goal-driven assembly tasks without relying on expert demonstrations or explicit description of the whole assembly. The proposed approach is validated through simulation and physical setup.