Action determination method and device, readable medium and electronic equipment

Abstract

The invention relates to an action determination method and device, a readable medium and electronic equipment, and relates to the technical field of machine learning, and the method comprises the steps: obtaining an environment image of an environment where a target object is located and an object feature vector of each environment object in the environment image, determining the position information of each environment object in the environment image, and for each environment object, splicing the object feature vector of the environment object to a position corresponding to the position information of the environment object, and taking the spliced environment image as input of a pre-trained action model to obtain a target action to be executed by the target object output by the action model. According to the method, the object feature vectors are spliced to the positions corresponding to the position information, so that the object feature vectors and the position information have relevance, the action model can quickly extract important features from the spliced environment image, and the accuracy of the target action output by the action model is improved.

Description

technical field [0001] The present disclosure relates to the technical field of machine learning, and in particular, relates to an action determination method, device, readable medium and electronic equipment. Background technique [0002] In recent years, DRL (English: Deep Reinforcement Learning, Chinese: Deep Reinforcement Learning), as one of the important research directions of machine learning technology, has been widely used in the fields of games, robot control and automatic driving. DRL can be divided into two stages: environmental cognition and decision-making. In the environmental cognition stage, the deep neural network is usually used to perform representation learning on the multi-modal fusion environmental information, and then output the deep semantic information obtained by representation learning to the decision-making network for decision-making. learn and generate corresponding optimal actions. [0003] Usually, when the environment information includes ...

AI Technical Summary

Problems solved by technology

However, using such an information fusion method cannot reflect the correlation between the environment image and object information, which will make it difficult for the DRL algorithm to extract important features from the environment information, and thus make it difficult for the DRL algorithm to generate correct actions.

Images