The Research Of Multipath Planning Based On Evolutionary Neural Network In Unknown Environment

Path planning of mobile robot is one of key issues in robotics research, and also is a focus topic for the researchers all over the world. At present, considerable achievements and successful application have been made in mobile robot path planning based on known environment; some achievements have been made in mobile robot path planning based on partly known environment or unknown environment, but there are many problems. The task of current mobile robot path planning is mainly to find only one optimal path. But in exploration of deep sea and space, because of uncertainty of the working environment, only an optimum or optimal path can not meet the requirements in practice. Multiple optimal paths are planned necessarily and right path can be selected temporarily according to different needs of task.Aiming at these problems, following research is made in this paper:(1) As the relevance between multipath planning and problem of multimodal function, neural network optimized by adaptive niche GA is constructed to solve problems of multimodal function. The evolutionary neural network constructed in the paper has the ability to find multiple peaks, and it is verified by the simulation of some multimodal functions.(2) Various methods about single-path planning in unknown environment and multipath planning have been analyzed. Aiming at demands of multipath planning in unknown environment, a method that uses information of robot's location and obstacles detected by sensors is construted to distinguish different feasible path in unknown environment. Based on this, a neural network optimized by adaptive niche GA is proposed to apply in unknown environment. The diversity of population is maitained in the process of evolution, so different feasible paths can be searched by robot, and then multipath planning can be realized. Then simulation is carried out: information of local environment is detected by sensors, and the robot's next move is determined by its "brain": neural network. Through constantly "learning" in environment for a time, robot can avoid obstacles effectively and reach the target point safely, and multiple optimal paths searched are smooth.