Research On Localization Techniques For Search Robots

Localization for robots is always the keystone and difficulty, and it provides information for autonomous navigation. It is significance to improve performances of robots in intelligence. There are several methods of localization for mobile robots. However, literatures are rare about localization for search robots in catastrophic coal mines which are representations of closed and unconstructed environments. Many challenges must be confronted when people develop localization systems for search robots. For instance, both methods of localization and selection of sensors are restricted. Basd on sensor techniques, signal filtering algorithms, multisensor data fusion techniques, this dissertation researched on integrated localization techniques for search robots used in closed and unconstructed environments and proposed two localization systems for search robots. It also researches on methods to compensate and modify errors during localization.In this thesis, two integrated localization methods are proposed. One is a strapdown inertial integrated localization system mainly based on inertial sensors and a digital compass; another is a combination of optic encoders and a digital compass. A hardware platform for a strapdown integrated localization system based on inertial sensors isdesigned and manufactured.Several main factors are analyzed which affect validity of data from sensors for a strap-down integrated localization system. A wavelet filtering algorithm based on optimal algorithms is proposed. Because parameters including wavelet function, decomposition scales and filtering thresholds in each scale have significant influences on effect of wavelet denoising. Both power spectrum density of surrogates reconstructed by certain frequency coverage and computational cost are taken as synthetical index during selection of wavelet function and decomposition scales. It provides an effective method for selection of wavelet function and decomposition scales. And then, a method setting filtering thresholds in each scale is proposed. It is based on a fact that road irregulartie results in many disturbances. Therefore, properties both sensors' noises and those disturbances should be taken into account in order to calculate filtering thresholds. A semi-physical simulation has been done.Two proposed integrated localization algorithms are put forward in this dissertation. One is based on a strapdown inertial localization algorithm and a multisensor data fusion method; another is designed for a localization system integrating optic encoders and digital compass. Sensors in the strapdown localization system provide data and federate information fusion algorithm is a theoretic foundation. This thesis brings forward a hybrid federate filter which combines feedback-fusion mode and mode without feedback to meet requirements for strap-down localizations. Unscented Kalman filtering algorithm is the kernel of the hybrid filter. It acquires a balance between precision and fault tolerance for integrated localization systems. A method for selection of filtering thresholds was proposed on the base of filtering algorithm for sensors outputs, which extended range of robots application.A localization algorithm is developed for systems based on optic encoders and digital compass. This algorithm can avoid calculation of heading by means of information from two encoders in order to decrease accumulative errors of attitude during dead reckoning. It can position robots in three-dimensional space.Corresponding error models for two localization methods have been founded from error sources. And then, measures are mentioned to compensate and modify errors. Those error models disclose rules of error transmission and provide theoretic foundation for compensation. Some methods are proposed to modify and control errors on the basis of working pattern of robots, which utilize information of maximum speed, nature landmarks in environments and identification situations of robots.Localization systems are designed and manufactured, a series of simulations and experiments were made in this thesis. Based on a platform for a strapdown integrated localization system, semi-physical simulations including signal processing and localization have been carried out. Data from sensors are processed by wavelet denoising algorithm based on optimal parameters and they are utilized to reckon tracks of robots during simulations. Outdoor experiments were made to position a tracking robot by means of two optic encoders and a digital compass. All of results show that methods of proposed localization in this dissertation can position robots without empirical information from environments and that they are suitable for tasks of searching victims and detecting dangers when they are used by tracking robots in catastrophic coal mines. A new idea and approach can be obtained to position robots in underground unconstructed environments from achievements in this dissertation.