Research On Swarm Intelligence Algorithm Based On Robot Swarm Learning

In recent years,with the rapid development of technologies have ushered in an inflection point in development,such as big data,artificial intelligence.Artificial intelligence will become the most critical technology in the world in the next 20 years,and will become an important foundation for emerging industries such as robots,drones,and smart companionship.At the same time,as a branch of artificial intelligence,advances in machine learning will further promote the development of artificial intelligence.We began to believe that artificial intelligence can make human life better.The behavior of robots in the context of artificial intelligence and machine learning can organize themselves to form groups to perform complex tasks.This has attracted wide attention in the academia.In this group,each robot can freely form a group with any other robot and exchange information.In the early 1990 s,the group intelligence algorithm was first proposed.Since then,many different swarm intelligence algorithms have been proposed.More and more sophisticated intelligence algorithms are used to better control the robot population,but few research has been devoted to describing different swarm intelligence algorithms and evaluating their performance in robot swarm learning to select more suitable swarm intelligence optimization algorithms for different robot populations.The Webots mobile robot simulation system can be used to model,program and simulate simulations.In Webots,six programming languages can be used to view and run simulation simulation programs in the three-dimensional simulation interface.It has the advantages of easy creation of environment,realistic simulation effects and direct familiarity with the compiled language Webots is very prominent in the study of the behavior of robots.However,at present,there are few researches in the Webots system to solve complex tasks by compiling controllers.In response to the issues,we conducted the following research: First,we selected three representative swarm intelligence algorithms based on different learning strategies of different swarm intelligence algorithms: Bat Algorithms with inexperienced learning ability and social learning ability,Particle Swarm Optimization Algorithms with self-cognitive learning ability and social learning ability,and Wolf Optimization Algorithms with social learning ability,Then,we summarize the flow and application of swarm intelligence algorithms.For the problems of their precocity and local optimism,we added Differential Evolution Algorithm in the individual updating process to improve the individual differences and ensure the global search ability.Afterwards,the Khepera III robot was applied to the Webots mobile robot simulation system,and the simulation environment was created by adding and modifying the controls and parameters.According to the behavior of the robot to avoid obstacles,the distance between the obstacle and the obstacle is kept as far as possible with the fastest straight line speed and the minimum degree of turn.A mathematical model for simulation based on intelligent algorithm is established.Using the characteristics of the information exchange between the robot controllers and the supervisor,by compiling the robot controllers and supervisor based on the artificial neural network,the problem of solving the model is transformed into the maximum objective function of the spatial search.Finally,aiming at different swarm intelligent algorithms in robot swarm learning obstacle avoidance behavior problems in different environments,we designed 27 sets of different experiments and analyzed the experimental results as follows: 1)Comparison and analysis of optimization process and optimization result of three swarm intelligent algorithms;2)Comparison and analysis of the performance of the three swarm intelligence algorithms under different parameter settings;3)Comparison and analysis of the performance of each algorithm under different NR;4)Comparison and analysis of the performance of each algorithm under different CR.The following research results have been obtained through group comparison of box graphs and verification of t-test statistics 1)Compared with BA algorithm and GWO algorithm,in solving the problem that intelligent groups get obstacle avoidance ability through learning,the PSO algorithm with relatively outstanding performance may be prioritized,for example,in the obstacle avoidance problem of cleaning robots;2)When the number of smart groups is large and the communication range is large,the GWO algorithm can be applied,for example,in a robot soccer game problem.3)Group intelligent algorithms that only have social learning ability,such as GWO algorithm,improving social learning ability will definitely bring better results,but swarm intelligence algorithms with different learning ability,such as BA algorithm and PSO algorithm,improving social learning does not necessarily improve the optimization results.