Research On Learning Real-Time Path Planning Algorithm Of CGFs In Unit Tactics

With the development of computer simulation technology,the use of simulation technology to assist the military command training,can effectively improve the training organization efficiency and training effect.Under the conditions of informatization,more attention is being paid to the synthetic,miniaturized and versatile combat force,and the unit is playing a prominent role as the main combat unit,so the construction of the combat capability of the unit is an inevitable trend.Conducting research and application of unit-level training simulation systems can efficaciously tackle the problems of traditional training approaches in terms of organizing,coordinating,the construction of battlefield environment and the feedback of combat effect,which possesses considerable significance.In order to promote the realism and application effect of unit-level training simulation systems,it is crucial to build reasonable battlefield space representation and accurate behavior model of Computer-Generated Forces(CGF).In unit tactics,path planning is an important intelligent behavior.It is not only the basis of many high-level behaviors(e.g.mission planning and coordination),also the pre-condition of a multitude of fundamental physical behaviors(e.g.maneuvering,covering and formation control).Therefore,rational and accurate path planning has been one of the focuses of CGF behavior modeling.Conventional planning methods(like A*),in which first-move delay obviously exists,cannot quickly respond to user commands and environmental changes.Conversely,LRTS(Learning Real-Time Search),a typical local planning approach,interleaves planning,learning and executing the movement in every single iteration,only with locally perceived information.Thereby,it is a method whereby to dynamically adapt to the changing environment and resolve multi-agent path planning.However,with over-optimistic evaluation for terrains thus inaccurate heuristics,LRTS learns slowly in complicate terrains and scrubs irrationally,which collapses the behavioral realism.In order to address above problems and accelerate the learning process of LRTS,this paper proposes a real-time search algorithm EQ-LRTS based on a multi-resolution quad-tree model using state aggregation and map hierarchical abstraction.Firstly,the non-pointer coding method based on bit operation is designed,and then the recursive construction process of quad-tree and the local repair mechanism of the model triggered by environmental changes are established.Based on the above method,the incremental environmental model and the planning and learning process are set up.EQ-LRTS shows the outstanding dynamic adaptability given the initial inaccurate map,and gradually updates and refines the environmental model as CGF explores.In addition,compared to the traditional grid discretization with fixed size,the multi-resolution characteristics of the quad-tree can effectively improve the planning efficiency.EQ-LRTS improves the planning and learning efficiency of plain LRTS in general scenarios,but CGF still scrub in the local minimal region of the heuristic.In order to solve the above problems and improve the rationality of the algorithm,this paper simulates the routing process of human beings under the condition of limited perceptual range,and proposes a hybrid algorithm of boundary following and LRTS,namely BF-LRTS(Boundary Following based LRTS).The application of boundary following method in real-time path planning is studied.Then,the whole process of hybrid algorithm is proposed,and the switching conditions and control strategies between different behavior patterns are designed according to local terrain features.This method has achieved very significant effect on performance indicators such as state re-visitation,which can effectively improve the path rationality,so that CGF can be quickly escape from the minimal area.The simulation results show that the multi-resolution terrain abstract model of EQ-LRTS can effectively improve the learning speed of the algorithm and reduce the computational complexity of the planning process,the number of state re-visits and the frequency of the update of the heuristic value.The hybrid algorithm BF-LRTS can greatly improve the real-time path planning rationality and realistic degree,to achieve system application requirements.