Method and device for actively constructing environment scene map by intelligent agent and exploration method

Abstract

The invention provides a method for actively constructing an environment scene map by an intelligent agent based on visual information, an environment exploration method and intelligent equipment. The method comprises the following steps: collecting an environment scene image required by a training model and a corresponding environment scene map data set; collecting an intelligent agent exploration environment path required by the training model; training an active exploration model by adopting an environment scene image, a corresponding environment scene map data set and a collected agent exploration environment path; and an action is generated based on the trained active exploration model, the intelligent agent explores the environment by adopting the generated action, 3D semantic point cloud data is obtained in the exploration process, and then an environment scene map is constructed by utilizing the 3D semantic point cloud data. According to the method, the limitation that a traditional computer vision task can only passively perceive the environment can be overcome, the active exploration characteristic of the intelligent agent is utilized, the perception ability and the motion ability are combined to achieve scene maps of active perception, active exploration of the environment and active construction of the environment, and the method is applied to various vision tasks.

Description

technical field [0001] The present invention generally relates to computer vision technology, and more specifically relates to a method, an intelligent device and an exploration method for actively constructing an environmental scene map by an intelligent body imitating a creature and a human. Background technique [0002] With the continuous development and wide application of machine learning technology, the field of computer vision has developed rapidly in recent years, and has achieved many remarkable results, including object detection, object recognition, and image semantic segmentation. However, these traditional 2D computer vision tasks can only passively perceive the environment and focus on analyzing the static environment, which ignores a very important feature of an agent (including creatures and humans): active exploration, that is, continuous Actively explore the environment to construct an environmental scene map and guide actions. Active exploration can prom...

AI Technical Summary

Problems solved by technology

However, these traditional 2D computer vision tasks can only passively perceive the environment and focus on analyzing the static environment, which ignores a very important feature of an agent (including creatures and humans): active exploration, that is, continuous Actively explore the environment to build an environmental scene map and guide actions

Images