Research On Cognitive Behavior For Two-wheeled Robots Based On IM-ELM Algorithm

In view of the problem of higher dimensions,slower training speed,easier to fall into the local optimum in traditional BP,SVM and other basic network which are applied to balance control and other cognitive behaviors of two-wheeled robot,an extreme learning machine has been proposed.Combined with intrinsic motivation and other different ways,it replaced intrinsic reward of robot with intrinsic motivation to improve learning efficiency and accuracy.The main research achievements are as follows:1)ELM algorithm based on intrinsic motivationIn view of the problem of the poor initiative in the balance of motion control of twowheeled self-balance robot,an ELM algorithm based on intrinsic motivation has been proposed.It replaced internal reward signal by intrinsic motivation mechanism to improve independent performance of learning and let robot learn balance skills gradually.2)Online sequential ELM algorithm based on Q learningIn view of the problem of poor adaptive learning ability in unknown environment,combined with reinforcement learning theory,an online sequential ELM algorithm based on Q learning has been proposed.It makes robot learn intrinsic relationship between the signal and behavior gradually and achieve the obstacle avoidance task in simple and complex unknown environments.3)ELM algorithm optimized by improved particle swarmDue to the random selection of weight and threshold parameters,it also proposes an ELM algorithm optimized by improved particle swarm.It is used to optimize the input weights and threshold of extreme learning machine to improve the structure of ELM.It could overcome inaccuracy of traditional extreme learning machine through particle swarm optimization algorithm.These algorithms have greatly improved the speed and accuracy of autonomous mental development of two-wheeled robots,which provide a new idea and method to improve the learning speed and precision of autonomous cognitive development of agents.