A Counting Method based on Deep Reinforcement Learning Combined with Generative Adversarial Network

In the process of function optimization and scheduling of multi-agent system model, the precise step calculation has attracted more and more attention. How to accurately calculate the number of steps has become an important research area, and it is also an important prerequisite for multi-agent cooperation in navigation and collision avoidance. This function can also be extended to intelligent wearable devices. The main drawbacks of the current popular methods are that they have extremely high requirements for noise data filtering, and the parameters in the model are not updated in the calculation process. To solve the above problems, this paper proposes a step counting algorithm of deep reinforcement learning combined with generative adversarial network. The critical contains a generator and a discriminator. The generator is used to learn the value function distribution, the generator and discriminator are learned at the same time with the help of confrontation training, and the generated state action values are used to train the policy network. The algorithm is applied to the step counting, which can further improve the accuracy of the step counting while accelerating the learning rate of the original gradient algorithm.