GP world model using strategy model to assist in training and training method thereof

Abstract

The invention discloses a GP world model utilizing a strategy model to assist training and a training method thereof, the GP world model comprises loss functions used for training the world model, the loss functions comprise a first loss function and a second loss function, the first loss function is an own loss function of the GP world model, and the second loss function is an own loss function of the GP world model. The second loss function is a loss function of the strategy model, and the training method comprises the following steps: S1, the world model updates model parameters by using the loss function; s2, updating model parameters by the strategy model by using a loss function, and storing each step in the training; and S3, taking the average value as a descendant for next training of the world model. According to the method, a training mechanism of a method for training the GP world model through the strategy model in an auxiliary mode is provided, the purpose of modulating and training the world model can be achieved through the stability of strategy training, and therefore the training effect and performance of the world model are improved.

Description

technical field [0001] The invention belongs to the technical field of world models, and in particular relates to a GP world model and a training method thereof using a strategy model for auxiliary training. Background technique [0002] The deep reinforcement learning framework is a framework that can well solve the problem of limited sample data. The deep reinforcement learning framework mainly includes two parts: the policy model and the world model. The policy model is trained using the experience in the experience pool. The world model imitates the environment by learning state transitions and rewards. The experience generated by the world model learning environment is also stored in the experience pool to provide more training data for the policy model, so it can overcome sample data. Insufficient problem. [0003] At present, the policy model and the world model of deep reinforcement learning are trained separately: the simulated experience generated by the world mod...

AI Technical Summary

Problems solved by technology

The applicant has found in the long-term research that the world model training effect of deep reinforcement learning implemented in this way is not good, but there has been no suitable solution before

Images