Affection Modeling And Behavior Decision-making For Multiple Affective Robots

Affective computing is getting more attention in the field of artificial intelligence.Behavior decision-making is an important research content of multi-robot cooperation system.Considering the effect of affection on behavior,the behavioral decision of affective robot in multi-robot cooperation system is a more interesting and complicated problem.Affection is a signal of communication,regulates social behavior,and renders robot and robot team more autonomous and efficient.The main research contents of this paper include:(1)An interpretable and computable affection model is proposed for the robot in multi-robot cooperation system.Combining with our previous fundamental work,we construct an affection model.The model includes several basic elements of personality,emotion and willingness,and several influences relationship of emotional decay,emotional contagion and external stimulation.This paper describes modules of the affection model and their coupling relationship in detail,proposes a new personality model named CASE to describe the traits of robot in cooperation,and optimizes the definition and calculation of stimuli.Willingness is proposed to measure robot's will to execute task,and is the main basis of task allocation.Finally,in simulation experiments,the positive effects of affective components in various aspects of cooperation are demonstrated.(2)Task allocation is one of the key issues of multi-robot cooperation,namely how to organize robots to execute corresponding sub-tasks efficiently and harmoniously.Based on the proposed affection model,a new task allocation algorithm for the multi-robot cooperation system with affection is designed,which try to exert the positive effects of affection.The effectiveness of this task allocation algorithm is verified by the pursuit-evasion game simulation experiments.(3)Behavior decision in the multi-robot cooperation system includes the behavior strategy in the task execution besides.Take pursuit-evasion game as instance of cooperative task,this work also studies the multi-robot pursuit strategy.The idea of potential point allocation is introduced,and extended the set of potential points to continuous space.The definition of continuous potential points is given by combining with the moving direction of evader.Pursuers choose different pursuits by allocating different potential points to pursuers.The deep deterministic policy gradient algorithm is used to optimize potential allocation strategy of pursuit team,and converge throughmultiple iterations.Finally,the effective convergence of the deep deterministic policy gradient algorithm is demonstrated.The intelligence and efficiency of the pursuit algorithm based on potential points allocation are verified,and the advantages and disadvantages of the algorithm are analyzed.