Dead reckoning method has significant drawback that it can't detect slip, unexpected material and base unflatness while it is broadly used for position estimation of mobile robot because of its simplicity, easiness and fastness in algorithm. This paper presents neural network based position estimation method in slippery environment as an approach to solve one of problems which are engaged in dead reckoning method.
Position estimator is composed of slip detector and linear velocity estimator. Both of them are based on dynamic characteristics of mobile robot. Slip detector is developed using the difference of dynamic characteristics between in slip and in no slip condition. Linear velocity estimator in slippery environment is also developed using dynamic relation between driving torque, angular acceleration of driving wheel and linear acceleration of mobile robot. To find out the relation, accelerometer is used for measuring acceleration of mobile robot and neural network is used for dynamic system identifier in slippery environment. A lot of experiments are performed to make data to train neural network with various driving condition.
After neural network training, a series of pathes including slip region are tested for position estimation and path tracking to evaluate the performance of position estimator.
