Research On Key Technologies For Combat Training Robot

Combat training is the main way of military training in today's peace age.Traditional training method can not satisfy the requirement of training innovation,because the limitation of technical conditions and cognitive level.This paper focus on the research of object recognition algorithm,path planning in dynamic surroundings and cooperative learning method for combat training robot.The main contents of this paper are as follow:(1)According to the characteristic of the time-varying of target location in combat training environment,a novel deep convolutional neural network based on optical flow(OF R-CNN)is proposed.The mobile objects in image are extracted as candidate regions with improved optical flow,the object recognition is carried out by using deep convolutional neural network.It solves the problem of slow detection speed of traditional method caused by exhaustive search algorithm.Optical flow was improved by pyramid algorithm to increase the speeds of region proposal.For the problem of slow convergence speed of traditional deep convolutional neural network,optimized the deep convolutional neural network algorithm by batch normalization methods.Finally,the generality and effectivity of this algorithm are proved on large data sets.(2)The problem of path planning in dynamic surroundings for combat training robot is studied,a hybrid path planning method APF-PRM was proposed which combines global path planning and local path planning method.Firstly,a collision free path is planed by pobabilistic roadmap method and obtained the path point sets.The"auxiliary force" is generated by path point sets,which solves the problem of falling into local minima of artificial potential field.The proposed APF-PRM path planning method is optimized by particle swarm algorithm,in order to plan a more secure collision free path.Finally,simulation is performed and results show that the path planned by optimized APF-PRM is shorter and securer.(3)For the problem of combat training in volatile environment,the research of the algorithm of autonomic learning and intelligent decision algorithm for combat training robot was carried out.Firstly,the characteristics of the combat training are summarized,and according to these features,a novel reinforcement learning system for combat training robot is proposed based on multi-agent reinforcement learning theory.At last the simulation experiment proves the feasibility of the proposed algorithm.(4)For the curse of dimensionality in Team Q-learning algorithm,TQ-Kmeans based on K-means with variable resolution was proposed,the similar states classified as the same categories by calculating similarity of environmental states set,whitch significantly reduce the number of environmental states.Finally,simulation is performed and results show that TQ-Kmeans algorithm has a faster convergence speed.This paper developed the simulation experiment platform of combat training for mobile robots based on Python,study results are applied to the combat training system,simulation results proved the feasibility and validity of the proposed algorithms.The research of this paper is significant to the development of combining artificial intelligence technology with military training.