Design And Implementation Of Intelligent Command And Control System For Multiple Moving Objects

With the rapid development and application of artificial intelligence,on the one hand,the burden and pressure faced by front-line workers is reduced,on the other hand,people also hope that artificial intelligence can help with the decision making process from a higher level,e.g.providing references,suggestions or even making decisions directly.The biggest difference between the intelligent command system and the traditional coordinated control system is that the former has higher real-time requirements.It needs to predict the situations that may occur during the work of multiple controlled objects,and to deal with unexpected situations in real time.In the path coordination task of multiple moving objects,the intelligent command system needs to be able to make real-time decisions and give control instructions according to the current state.To meet those requirements mentioned above,this paper designs and implements an intelligent command and control system that be able to coordinate multiple moving objects in real time.The system is composed of a human-computer interaction module,a simulation running module,a command coordination module,an agent training module,a motion plan database and an agent model database.Among them,the simulation running module and the command coordination module are the cores for command and control function and the agent training module is used to train agents.The simulation running module provides location information based on real-time scene graphs for users,and performs collision detection of moving objects based on path checkerboard graphs.In order to prevent and avoid command deadlocks in the system,this paper designs and adds a deadlock detection and prevention strategy.In terms of the high reliability requirements of collision avoidance in real-time command scenarios,the command coordination module uses a collision prevention strategy to screen out the actions that may cause collisions and then the reinforcement learning agent determines the action instructions to reduce task completion time.This article uses the Py Charm platform and the Py Qt framework to develop various parts of the system which is used as an environment to experiment the agent training algorithm.The agent training module is used to train agents for different command tasks.In response to the flexibility requirements of scene task setting,the corresponding random sampling strategy and reward model is designed.This paper designs and implements the agent training algorithm based on the DDQN and the PPO algorithms respectively.The agents trained by these two algorithms can be loaded and complete real-time command tasks.The collision prevention strategy helps the training algorithms reduce the computational overhead,and get a better training effect and convergence speed.In order to prove the effectiveness of the reinforcement learning algorithms,the greedy strategy algorithm is added for comparison.In different task scenarios,this paper designs and conducts corresponding comparison experiments.The results show that both reinforcement learning algorithms have good coordination effect.