Study Of The Evolution Of Robot Behavior Based On Neural Network Ensemble

Evolutionary Robot (ER) is the robot using evolutionary algorithm and has the ability to adapt to dynamic environment. It can solve the above problem well. A new parallel evolutionary algorithm based on neural network (NN) Ensemble is proposed for mobile robots. The robot controller has a reconfigurable structure with user-defined sensor suite, which corresponds to a typical network. NN Ensemble is evolved to optimize the optimal sub-networks including both learning rules and combining weights dynamically. The experiment results show that the genetic algorithm based NN Ensemble has stronger evolutionary capacity and higher learning efficiency of parallelism compared with the general genetic algorithm.The paper discusses the hardware implementation of genetic algorithm. A hardware architecture based on FPGA (Field Programmable Gate Array) is presented, which is capable of population storage, selection, crossover, mutation and fitness evaluation. The design utilizes Altera FPGA Cyclone EP1C3T144C, which shortens processing time and has broad application.At last we design the simulation system using Object Oriented programming. The simulation environment is based on C++Builder platform. Model of sensors and the method of map building have been discussed. According to the evolution result, it has been proved the validity of evolution algorithm.We hope to simulate the human' s study and evolution to research the behavior learning and evolution of autonomous mobile robot. In this system, we only program some based behavior and action modes of the robot. The robot can learn to modify these action codes through moving in the environment autonomously. The method can evolution the robots' performance and behavior through the normal genetic arithmetic to make the robots adapt to the dynamic, unknown and unstructured environment. The methods can let the designer be far away from designing some basic actions of the robot and makes up some deficiency of autonomous mobile robot, increase robots' practicability.