Research Of Path Planning Based On Evolutionary Spiking Neural Networks

Some success has been achieved in the study of autonomous agents for pathplanning process based on neural driven. Currently, certain problems still exist in theevolutionary optimization of autonomous agents neurocontroller. In view of this,based on spiking neural networks this paper proposed an evolutionary autonomousagents path planning method, which uses a more bio-interpretative developmentalapproach and evolutionary mechanism, uses spiking neurons in more line with theBiological characteristics and more bio-interpretative learning rule to optimize thedesign of neural network controller.By adopting the dynamic characteristic of recursive genetic regulatory networksto show the development process of specialized cell fate, the method applies weightreplacement operator and additive noise operator to mute recursive genetic regulatorynetworks. We perform path planning experiments on the spiking neurocontrollerthrough the development and evolution of Autonomous agents both in static anddynamic environment. And it is also tested the influence of path planning by life-timelearning.In the static environmental experiment, we firstly verify the path planning resultsof the different sensorimotor systems and autonomous agents of scales of the spikingneurocontroller. Secondly, we analyze the impacts on path planning of autonomousagents by changing the regulation of genetic regulatory network nodes and regulatoryweights scale. Then, we use weight replacement operator and additive noise operatorin the variant process of recursive genetic regulatory networks respectively andanalyze the impact on path planning of autonomous agents brought by the twodifferent mutation operators by comparison. Finally, we verify the performance ofpath planning of autonomous agents by changing their sensing radius and runningspeed.In the dynamic environmental experiment, we firstly verify the path planningresults of different sensorimotor systems and autonomous agents of different scales ofthe spiking neurocontroller and make a comparison with that in static environment.Secondly, we analyze impacts on path planning of autonomous agents by changingthe number and speed of obstacles in natural field. Based on the importance of synaptic plasticity in the biological life-time learning,the learning rules on the spike-timing-dependent plasticity (STDP) and correspondingmathematical models are applied. And experiments on different scales of the action ofspiking neurocontroller for autonomous agents path planning also implemented bothin static and dynamic environment.Simulation results show that the path planning method based on evolutionaryspiking neural networks provides better timeliness and adaptability for autonomousagents both in static and dynamic environment, which can reach accurate positioningof the targets.