Support vector machine based artificial potential field for autonomous guided vehicle

Abstract

The path planning is developed on the subject which aim to guide a walking robot from a starting point forwards a goal. The paper presents a now model merging the optimization of support vector machine (SVM) into the artificial potential field path planning. Using the path planning, robots can estimate a free smooth walking path of obstacles. Based on the statistical learning theory, the SVM can be used to optimize a zero-potential decision boundary in the 2-dimemsional map with a large margin. The idea of large margin implies that a wide path can be obtained with the employment of the SVM. With the RBF kernel, the presented method produces a 2-dimemsional potential-field map. In the map, obstacles are modeled as the sum of various parametric Gaussian distributions. As known, a map composed with the superposition of 2-dimemsional smooth Gaussian functions can also achieve the walking path smooth. Upon this, potential-field or road map based robot navigation can easily be applied to achieve the path smoother. The proposed model provides a way to search a wide smooth road for the robot. Detailed experiments and discussions are included in the paper.