Research On Autonomous Mapping Of Mobile Robot

Autonomous mapping has been a highly active research area in the field of mobile robots during the last decade. As one of the key techniques to build truly autonomous mobile robots, it can greatly enhance mobile robots'ability of interacting with environments, and therefore favors the development of robotics. Virtually, it will even deeply influence the existence and development of human beings because mobile robots with the technique of autonomous mapping can replace the place of human beings to be applied on various situations, for example, outdoor or sea bed projects, disaster rescue and planetary exploration. Despite significant progress in this area, it still pose great challenges, such as appropriate map representation, reducing computational complexity and mapping in real time, improving the accuracy of the estimated map, mapping in dynamic large scale environments and so on. This thesis studies the problem of autonomous mapping with mobile robots and the major results of the research are described as below:1. The background of mobile robots such as category, history and research subjects is briefly reviewed. Then the problem of autonomous mapping is divided into three sub-problems from the easiest to the hardest as obstacle avoidance and wandering, localization and SLAM (Simultaneous Localization and Mapping). Finally, the theories and techniques to these problems are systematically summarized.2. Based on the concept of traversable corner and un-traversable corner, an un-calibrated monocular visual wandering algorithm with the piecewise control law is proposed according to the characteristics of the static environment consisting of right angles and straight lines. With the algorithm, a mobile robot equipped with an un-calibrated camera can wander in the environment or even reconstruct the topological map with the recorded corner information. The experimental results prove the effectiveness of the proposed algorithm.3. Distance and angle control model has the advantage of platform free. To build the map of static structured environment, a method which takes it as the motion model under the framework of the SLAM using Kalman filter algorithm (EKF-SLAM) is proposed. In order to reduce the amount of calculation, a new environment description method combining point landmarks and its adjacent matrix which is calculated with the landmarks and its vectors is put forward. The simulation and experimental results indicate the proposed algorithm could build accurate maps whose information is complete.4. The mapping accuracy and efficiency are incompatible since the accuracy of the map generated with the EKF-SLAM algorithm can be improved only if the mobile robot keeps on moving around and taking more measurements without condition changes. A map improvement method therefore is proposed. It constructs constraints with prior heading and measurement information according to the covariance matrix and applies them to the estimated map. The algorithm is tested in simulation with three different kinds of constraints. The results show that the constraint constructed with the maximum variance and minimum variance landmark pair can greatly improve the map accuracy and balance the mapping accuracy and efficiency. The experimental results also prove the effectiveness of the algorithm.5. The FastSLAM algorithm based on the particle filtering has to resample to deal with the degeneration. But resampling will also lead to particle depletion and ruin landmark estimate diversity because the characteristics of time forgetting is not fulfilled for the historical estimate error is coded in the maps. So resampling should be avoided if unnecessary. A new particle weight calculation method based on smoothing which could suppress the excessive fluctuation of the particle weights and reduce resampling frequency is proposed. In the method, particle weight is codetermined by the instant information and its history weights recorded in a dynamic sliding window. Simulation results demonstrate that the proposed method can result in more accurate maps.6. A novel dynamic environment model is proposed, which combines the occupancy grid model and sample-based model.In order to take its advantage of information fusion, the occupancy grid model describes the static parts of the environment. The sample-based model which is composed of two sample sets describes the dynamic objects in the environment because it needs no prior information about the dynamic objects such as position, distribution and motion model. The first sample set keeps and updates the latest samples of dynamic objects and is benefit to obstacle avoidance. The second one keeps all the samples of the dynamic objects in the history. With proper postprocess, it can provide active regions of the dynamic objects which are critical complement to the occupancy grid map. The experimental results show the resulting map is fine and useful for path planning and navigation.