Robot behavior teaching method based on meta-learning

Abstract

The invention provides a robot behavior teaching method based on meta-learning. The robot behavior teaching method is characterized by comprising the steps: acquiring a teaching video; and learning the teaching video by using the trained neural network model. The training process of the neural network model comprises the following steps: collecting training content; preprocessing the training content to obtain a preprocessed comparison video, a preprocessed teaching video and a preprocessed motion video; constructing an initial neural network model; taking the preprocessed teaching video as aninput, obtaining a demonstration action, and calculating the loss of the demonstration action; updating the initial neural network model according to the demonstration action loss to obtain an updated model; taking the preprocessed motion video and the track action as input to obtain a predicted track action, demonstrating semantics, motion semantics and comparison semantics, and calculating target action loss and semantic loss so as to construct total loss; updating the updated model based on the total loss; and until the total loss is stably converged to a total loss threshold, obtaining atrained neural network model.

Description

technical field [0001] The invention relates to the technical field of program-controlled manipulators, in particular to a robot behavior teaching method based on meta-learning. Background technique [0002] With the development of artificial intelligence technology and the wide application of robots in aerospace, education, service, detection and medical fields, the intelligence of robots has attracted widespread attention. However, most traditional robots have a limited level of intelligence, lack of learning ability and ability to flexibly adapt to task changes. Taking industrial robots as an example, every time an assembly line is set up for production, it needs to be pre-calibrated and programmed by professionals, which is very troublesome. In practical industrial applications, it is often necessary to reset the production line due to business changes. Every time the production line is changed, it is troublesome to accurately locate the position of the workpiece and r...

AI Technical Summary

Problems solved by technology

Although meta-learning algorithms such as MAML algorithm perform well in the fields of regression, classification, image super-resolution and reinforcement learning, there are still many problems in the application process of meta-learning algorithms such as MAML algorithm.

For example, the majority meta-learning method [1,2,3,4] The performance of the algorithm degrades rapidly when only visual information is provided as a teaching during use

Images