Study On The Vision-based Detection And Planning Techniques For Hexapod Walking Robot Under Environment With Obstacles

Due to excellent static stability and adaptability to rugged environment or weak soil, hexapod walking robot is considered as ideal platform to execute special tasks including dangerous site exploration, disactrous accident rescue, humanitarian demining, et al. From the late20th century to these years, scholars from various countries have kept on researching on hexapod robots. Relatively mature kinematics analysis methods, mainstream dynamics approaches and hierarchical control strategy have been developed. However, there are still many problems in the tasks to make hexpaod robots more intelligent and autonomous, especially the rapid environment modelling, safe path and foothold planning in navigation.In the first chapter, after the review of domestic and oversea researches on hexpaod robot, the related key techniques and important issues are concluded. Based on the comparison of commonly used sensor techniques for environment exploration, the current status of target and obstacle detecting using binocular stereo vision is intensively focused on. The developing trends of existing coping strategy to obstacles and path planning methods are concluded, especially those for hexpod walking robots. Moreover, the progress of foothold planning and related representative works are analysed. Finally, several problems to be studied or adrresed in this dissertation are raised.The stereo vision based detecting of obstacles in sparse texture background is studied in the second chapter. Matching methods based on window correlation rely on abundant texture fetures and need to perform mass convolution, so real-time processing without dedicated acceleration circuit is impossible under high image resolution and big parallax searching range. Therefore, a solution is proposed to estimate the depth of obstacles based on block segmentation using image pairs without distortion rectification. The methodical error is analysed and a hyperthetical approximation model is raised. The parameter regression and validation experiments proved the feasibility of the proposed depth estimation approach.The definition of obstacle and coping strategies with application condition are given in the third chapter. Against the static and dynimic stepping over problems, mathematic relations between foot-end trajectory and obstacle overall dimension are built, the candidate foot-end trajectories for obstacle stepping over are compared through numerical analysis. Periodic and non-periodic foot placements are proposed for complete robot unilateral stepping over obstacles. The constraints between foot-end trajectory, body horizontal displacement during supporting phase and interval between hip joints are defined. Vision-based obatacle pose estimation method is given for robot course adjustment before bilateral stepping over obstacles. Measuring method of overall dimensions for obstacles in sparse texture background is given based on stereo vision. The random errors caused by image feature extraction are obviously depressed by using Kalman filter in data updating.Based on the movement characteristics and sensor configuration of the hexapod robot in this work, the local map generating method is given in the fourth chapter. The calculation methods for grid map parameters and obstacle information decomposition are described. The global map only records the position and orientation of obstacles and the robot. The updating algorithms for global robot information and local obstacle data are described, including the discrimination and register of obstacles observed at different moments. Dead reckoning based and binocular vision based algorithms are proposed for the self-localization of hexapod robot. Moreover, the integrated localization algorithm is given based on extended Kalman filter. The results of position and pose estimation experiments using reference blocks validated the proposed estimation methods for obstacles and robot body.In the fifth chapter, related issues for hexapod robot walking in scenes with traversable obstacles are discussed. The parameters needed for local path planning are defined and the simulation result with specific values of them is given. The global path planning includes two steps. First, a preliminary path is generated using potential grid method combining A*algorithm. Then path modification is executed to reflect the influence from two types of traversable obstacle. Particle swarm algorithm is adopted for candidate footholds searching in rugged environment. The specially designed fitness function integrated the purposes of path cost evaluation and the avoidance of forbidden zones. Through border erosion to grid map, the footholds close to zone border are decreased in the final planning result.