A syntactic method for robot imitation learning of complex sequence task

Abstract

Purpose Aiming at the problems of weak generalization of robot imitation learning methods and higher accuracy requirements of low-level detectors, this study aims to propose an imitation learning method based on structural grammar. Design/methodology/approach The paper proposes a hybrid training model based on artificial immune algorithm and the Baum–Welch algorithm to extract the action information of the demonstration activity to form the {action-object} sequence and extract the symbol description of the scene to form the symbol primitives sequence. Then, probabilistic context-free grammar is used to characterize and manipulate these sequences to form a grammar space. Minimum description length criteria are used to evaluate the quality of the grammar in the grammar space, and the improved beam search algorithm is used to find the optimal grammar. Findings It is found that the obtained general structure can parse the symbol primitive sequence containing noise and obtain the correct sequence, thereby guiding the robot to perform more complex and higher-order demonstration tasks. Practical implications Using this strategy, the robot completes the fourth-order Hanoi tower task has been verified. Originality/value An imitation learning method for robots based on structural grammar is first proposed. The experimental results show that the method has strong generalization ability and good anti-interference performance.