Research On Active SLAM And Related Problems In Human - Computer Interaction Environment

Intelligent mobile robots have been widely used in various aspects, such as home service, industry, exploration and military. The Simultaneous Localization and Mapping (SLAM) problem of mobile robot aims at:in unknown environment, the robot is able to build the accurate map of the environment, and simultaneously locate itself in the built map. In the traditional SLAM researches the robot moves by following a predefined path, while the Active SLAM research adds autonomous path planning to traditional SLAM, which gives the robot the ability of full autonomy. Therefore, the research of Active SLAM problem is of great significance, and has received extensive attention.The current researches of Active SLAM focus on static environment, lacking the recognition and tracking of dynamic objects. However, the real world is dynamic, and for service robots human-robot interaction is also inevitable. In this dissertation, the research of Active SLAM under human-robot interaction conditions is proposed. The purpose of the research is:when being guided to explore an unknown environment, mobile robot can perform intelligent path planning for SLAM, and track the guiding person simultaneously. The Active SLAM method we proposed use laser scanner to detect and map the environment, and use microphone array based sound source tracking to track the guiding person. The main works and novelty of the dissertation are:1. A prediction-based feature extraction algorithm is proposed to extract the geometric features such as lines and circles in the environment. The proposed algorithm simulate the process of laser scanning, and use the adjacent several data points to calculate the predicted position of the current point. By comparing the positions of prediction points and true points, the mark points of the raw scan data are detected. The proposed algorithm can separate all features in only one stage, while the traditional algorithms need two stages, thus the calculation speed and extraction accuracy are significantly improved. Moreover, the proposed algorithm does not rely on prior knowledge of scanner parameters, and the requirement of minimum point number of each feature is minimal.2. A hybrid sound source localization algorithm using microphone array is proposed. The traditional SRP-PHAT is accurate and robust, but due to full space search its speed cannot fulfill the real-time calculation requirement of mobile robot.The hybrid algorithm uses GCC-TDOA estimation to get the potential azimuths, and then calculates the narrowed search space by circular clustering. The SRP-PHAT is performed inside the narrowed search space, thus the speed is significantly increased. The hybrid algorithm can fulfill the real-time requirement of mobile robot applications, while maintaining the high accuracy and robustness of traditional SRP-PHAT.3. A fusing method to perform SLAM and Sound Source Tracking (SST) simultaneously is proposed. SLAM is based on EKF and laser features to get the estimations of system state and the corresponding covariance matrix. The sound source localization result is transformed from current local coordinates to global coordinates using pose information of the robot, and SST is performed through particle filtering in global coordinates. The method of SLAM-SST has not been proposed previously, and the method first proposed in this dissertation constructs foundation of Active SLAM under human-robot conditions. It will also provide new methods for mobile robot auditory research.4. The method of Active SLAM under human-robot interaction conditions is proposed, which expands Active SLAM research to dynamic environment. The optimal path planning problem when being guided is solved by optimizing local targets. The predicted position of guiding person, the pose of robot and the explored environment are all considered when generating potential targets and evaluating these targets. Laser scanner is utilized to perform SLAM, and microphone array is utilized to perform sound source tracking. Under this scheme, the robot is able to understand the intentions of the guiding person. Instead of simple following behavior, the robot can perform optimal exploration when being guided. The significance and value of the method proposed in this dissertation is verified by experiments in real environments.