Parameters Optimization Of CPG For Hexapod Robot Based On Multi-objective Genetic Algorithm

The traditional motion planning method is difficult to apply to multiple joints angle planning of hexapod robots.Reasonable angle planning is the basis for achieving stable motion.Generating the rhythm signal of planning robot motion in the analogy of the Central Pattern Generator(CPG)in creatures is an effective method to solve motion planning problems.For the hexapod robot which is studied in this paper,there are 49 parameters to be adjusted in the oscillator,which is constructed by the mathematical equations,the CPG formed by the oscillators,and the mapping function of the rhythm signals produced by CPG to joints' angles.Therefore,it is necessary to join the evolution optimization algorithm into the process of parameters tuning.Firstly,According to the mathematical model of Hopf oscillator,the influence of internal parameters on the output is analyzed.The central pattern generator is established by interconnecting Hopf oscillators.The influence of the coupling parameter on the output is analyzed.By mapping the state variable of the output of the oscillator to joints angle,the influence of the parameter in the mapping function on the output is analyzed.From the analysis above,the 9 independent parameters to be optimized is obtained,including intensity of convergence,amplitude,frequency,coupling strength and five scale factors.Secondly,the multi-objective optimization problem is mathematically described.In order to compare the relatonship between superiority and inferiority of the individual in the multi-objective problem,the Pareto dominance relationship is given.The definition of Pareto-optimal solution for the selection operation and the fast non-dominated sorting algorithm are also given.The exist problem and improving direction of NSGA-II(Non-dominated Sorting Genetic Algorithm II)are analyzed.The evolutionary cycle is divided into three periods:global convergence,diverse maintenance,and local convergence and the effectiveness of adjustment strategy is proved by the test function.Then,the simulation prototype model is built in the simulation environment.The sensor for collecting the motion states of the simulation prototype is selected according to the requirements of the optimization target parameters.The robot controller is constructed by the designed oscillator mathematical model,coupling relationship and mapping function.The monitor is constructed to reset the running environment of the simulation prototype.The data interaction between simulati on prototype and optimization algorithm in different processes is performed through the files.The data can be accurately interacted by signals synchronization.For there optimization objectives,including reciprocals of speed,setback and bump,the simulation optimization is perfermed by multi-object genetic algorithm and single object genetic algorithm.Finally,the data acquisition unit is built on the experimental prototype.The displacement information is collected by the distance sensor,and the speed is obtained by dividing the sampling time.The acceleration values of the forward direction and the vertical direction are obtained by the accelerometer.And the absolute values are integrated and averaged to represent the value of the setback and the bump,respectively.The compare between simulation results and experiment result illustrate the effectivnes of the above two optimization algorithms.In the experiment,the speed performance was improved by 11.63%.Both the setback and the bump are optimizated.