Data collection method and device for imitation learning and storage medium

Abstract

The invention discloses a data acquisition method and device for imitation learning and a storage medium. The method comprises the following steps: constructing an SLAM (simultaneous localization andmapping) map of an environment in which an imitation learning task is executed by an imitation object; obtaining a target path of the imitation learning task in the SLAM map; performing data acquisition based on the target path, wherein the data comprises image data and pose data of the imitation object.

Description

technical field [0001] The invention relates to data processing technology, in particular to a data collection method, device and storage medium for imitation learning. Background technique [0002] Imitation learning is a method for learning policies from expert demonstrations by cloning the expected behavior or understanding the rewards that the demonstrators are after. Imitation objects (such as robots) can achieve a demonstration effect close to that of an expert by simulating the demonstration. With the continuous improvement of the computing power of computing devices and the development of deep neural network technology, end-to-end imitation learning has made many progresses in the field of robotics (such as vision-based manipulator control, automatic driving), however, this end-to-end imitation learning A large number of expert demonstrations are required, and the quality of the demonstration data directly determines the upper bound of the effect of imitation learni...

AI Technical Summary

Problems solved by technology

With the continuous improvement of the computing power of computing devices and the development of deep neural network technology, end-to-end imitation learning has made many progresses in the field of robotics (such as vision-based manipulator control, automatic driving), however, this end-to-end imitation learning A large number of expert demonstrations are required, and the quality of the demonstration data directly determines the upper bound of the effect of imitation learning. There is no solution for efficiently and high-quality collection of data for imitation learning in related technologies

Images