Robot quasi-periodic motion demonstration learning method

Abstract

The present invention discloses a robot quasi-periodic motion demonstration learning method. The method employs classification, decomposition, modeling and synthetic technologies in order to achieve learning and generalization reproduction functions of robot quasi-periodic motion, namely, a classification method is employed to determine that a motion track is periodic motion, non-periodic motion or quasi-periodic motion; a decomposition algorithm is employed to decompose the quasi-periodic motion to the periodic motion and the non-periodic motion; the periodic motion and the non-periodic motion are subjected to modeling learning and prediction, and each component after learning is combined to a new quasi-periodic motion according to an equation of definition of the quasi-periodic motion. The robot quasi-periodic motion demonstration learning method successfully represents the complex motion track-quasi-periodic motion of the robot in a non-structural environment to effectively solve the technical problems that the class of flexible track traditional automatic planning system and method is single in function and bad in universality.

Description

technical field [0001] The invention relates to the technical field of robot motion trajectory demonstration learning, in particular to a robot quasi-periodic motion demonstration learning method. Background technique [0002] Learning from Demonstration (LfD) is a technology that imparts human skills to robots through human-computer interaction and teaching, which can greatly reduce programming and improve learning efficiency. The core of LfD is to perform data representation and learning on the trajectory of the robot. At present, most scholars are committed to the modeling and learning methods of the non-periodic motion trajectory of robots, such as artificial neural networks (ANNs) and hidden Markov models (Hidden Markov Model (HMM)) were used to solve the problem of perception in the early days. problems and reproducible problems, but these two methods require too much sample data to train an ideal model. Calinon uses a Gaussian Mixture Model (GMM) to encode the traje...

AI Technical Summary

Problems solved by technology

However, these systems lack universality, single function, and complex programming

In summary, the current demonstration learning methods and automatic trajectory planning systems cannot fully represent such complex motion trajectories.

Images