A Hierarchical Motion Retrieval Algorithm for Complex Manipulation Tasks Planning with An Encoded Knowledge Base

In human-robot cooperation, it is a challenge thing that the robot should perform to convert humans' natural languages to continuous action sequences, which is necessary for completing complex collaborative tasks. In this paper, firstly, a new knowledge base is built for encoding different features of movements, objects and relations; then, a hierarchical motion sequences retrieval algorithm is presented by combining our knowledge base with Deep Q-learning. Finally, the experiments verify that the developed reasoning system is effective and accomplishes to manipulate the objects to reach target statuses.