Robot imitation learning method based on Gaussian process

Abstract

The invention discloses a robot imitation learning method based on the Gaussian process. The method comprises the first step of arbitrarily setting a light source position; the second step of carrying out the action demonstration by a teaching robot to complete the phototaxis action; the third step of using a Gaussian process method to train a sample point set and establishing and solving a Gaussian process model; the fourth step of applying a mapping relation to a self control strategy by an imitation robot, and carrying out imitation learning of the action of the teaching robot; and the fifth step of carrying out imitation action analysis. The robot imitation learning method based on the Gaussian process applies the Gaussian process to research on the robot imitation learning control strategy algorithm. By collecting the teaching behavior sample points of the teaching robot, the Gaussian process algorithm is used to train the sample points, the mapping relation between the teaching robot's perception and behavior is learned, and the mapping relation is applied to the imitation learning control strategy of the imitation robot and then execution is conducted.

Description

technical field [0001] The invention relates to a robot imitation learning method, in particular to a robot imitation learning method based on a Gaussian process, and belongs to the technical field of intelligent products. Background technique [0002] Humans and other creatures can effectively acquire the movement patterns of their companions’ behaviors by observing and imitating the behaviors of their peers, and understand the meaning of these behaviors, giving robots the ability to imitate and learn like humans, and can effectively solve the problem of robot motor skills acquisition, Improving the learning efficiency and adaptive ability of robots is an important research direction of robot bionics research; the general learning form of imitation learning is action replication. The regression problem of executing the action is to determine the control strategy. The imitator executes the control strategy to reproduce the behavior and realizes imitation learning. Most of th...

AI Technical Summary

Problems solved by technology

[0002] By observing and imitating the behavior of their peers, humans and other creatures can effectively obtain the movement patterns of their peers and understand the meaning of these behaviors, endowing robots with the ability to imitate and learn like humans, which can effectively solve the problem of robot motor skill acquisition. Improving the learning efficiency and self-adaptive ability of robots is an important research direction of robot bionics research; the general learning form of imitation learning is action copying, which is mainly based on the movement trajectory of the teacher. The regression problem of executive action is to determine the control strategy. The imitator executes the control strategy to reproduce the behavior and realize the imitation learning. Most of the imitation learning algorithms use the inverse reinforcement learning algorithm to find the control strategy. The cost function of teaching behavior to imitate learning, and seek the control strategy that minimizes the cost function. However, the method of inverse reinforcement learning has high requirements on the cost function, and is not suitable for imitation learning tasks where the cost function is difficult to obtain