The demand of Auto Guidance Vehicles (AGVs) is increasing while the technique can be popularized to apply on industrial flexible path optimization in intelligent manufacturing and industry 4.0. In the general industrial environment, AGVs navigation is based on fixed guidance markers such as black line and QR code regarding to the cost-effectiveness. As deep learning technique carried out, Vision based Simultaneous Localization and Mapping (vSLAM) technique is widely used on unmanned vehicle auto-driving and navigation for the next generation of AGVs navigation. However, the challenges of vSLAM navigation technology are facing: 1) the vSLAM accuracy is up to 5~10cm, which cannot satisfy the parts of industrial requirements and 2) the problems of anti-disturbance ability are under complex environment, such as lighting, dynamic object, field of vision, therefore, those noise factors will affect the precision level of AGVs navigation. This paper focuses on the anti-disturbance problem of mobile robot's vSLAM navigation and finds out the solution. An optimization algorithm is developed to minimize robot path following error and enhance the performance of AGVs navigation. The algorithm is based on Lyapunov direct method controller with RBF neural network estimator.