Environment Map Building And Motion Planning Of Hexapod Robot Based On Binocular Vision

Compared to other types of mobile robots, hexapod robot has a strong ability to adapt to the ground and a good movement stability. It is beginning to show its advantanges in the fields of military reconnaissance, disaster relief, interplanetary exploration, blasting terrorism, archaeological detection, and so on. The ability of the complex environmental adaptability and automatic motion of hexapod robot are the key problems to be solved. Research will be resolved the issue of hexapod robot automatic motion in complex environmental as a starting point, focusing on three basic theoretical problems of real-time environment modeling, self-positioning and motion planning. And ultimately through a typical complex operating environment as demonstration, to verificate the adquired theory and technology innovations of improving the autonomous motion capabilities of the hexapod robot in complex environment.In order to solve the mutual restraint problem of the runtime and matching accuracy, a stereo matching algorithm based on support points is proposed. Firstly, epipolar rectification algorithm is applied to the the original image captured by cameras. Next step, using Canny edge operator to detect the edge point of left image as the stable support points. Divide and conquer algorithm is applied to achieve the 2D Delaunay Triangulation of the support points. Use triangular facets as matching primitives after the triangulation, establishing disparity model and energy minimization function to implement initial disparity estimation. Finally, according to the common characteristics of triangle share vertices with others to refine the initial disparity, obtaining the complete disparity map. After re-projection matrix to restore the three-dimensional scene information.A combination triangulation method of surface reconstruction based on Delaunay triangulation is proposed to achieve three-dimensional reconstruction of the point cloud. Providing front environmental information for robot motion planning. This combination triangulation method divided into three steps:(1) Divided the surface unorganized points by spatial bounding box algorithm, accelerate searching its K-Nearest Neighbor points. Constructor its tangent plan after point cloud data parameterized, simplify the initial point cloud data based on curvature can reduce the topology complexity of the point cloud effectively.(2) After the triangulation accomplish, according to the guidelines of opposite side disjoint rule, angle of two normal vectors maximized rule, semicircle distance rule and internal minimal angle maximized rule to implement Delaunay Triangulation in the area.(3)According to the criteria of the three-dimensional Delaunay empty external ball fusion in each sub-region to complete the final triangulation, and then achieve three-dimensional reconstruction of the scene. The combination of triangulation algorithm has the advantages of regional growth method and 3D Delaunay Traingulation, it can effectively reduce the topology complexity of the surface scattered points. It is suitable for closed and non-closed surface scattered points cloud triangulation. The triangle mesh is uniform, smooth, the detailed features of the surface can be better reflected. The runtime of the combination triangulation method is fastly, the runtime of the ten thousand scattered points triangulation is only 5.8s.Aiming at that the problem of low positioning accuracy and robustness only through binocular vision, the binocular vision combine with AHRS SLAM method of hexapod robot is proposed. Apply the GPU acceleration based SIFT algorithm to detect the matching feature points in the process of the robot motion. According to the known internal and external parameters of the camera, calculate the threedimensional coordinates of the point cloud in the robot coordinate system and build features maps. In order to prevent bad lighting of the scene or robot instability give rise to blur image, which then lead to visual odometry failure for the missing of visual information. Using the binocular vision sensor combined AHRS to completed SLAM of the six-legged robot. It can not only solve the inaccurate problem of feature point information only caputure by monocular vision using a special initialization, but also avoid the disadvantages of a large amount computing efforts by use single binocular vision to restore the robot motion. Experiments show that the algorithm is stable and high positioning accuracy in the indoor environment.Compared with wheeled, tracked mobile robot, hexapod robot has a more superior ability to deal with complex terrain. Indoor and outdoor unstructured terrain it is able to achieve stable walking and across the barrier. On the basis of the aforementioned acquisition of topographic information of the scene based on previous terrain imformation capture by binocular vision and terrain buliding, accomplish the foothold select according to the terrain geometry and the robot’s athletic ability Terrain geometry information mainly taking into account the rugged extent, height and the foothold stand area of the rugged terrain. Athletic ability of the robot mainly consideration the reachability of the robot leg and the stability of the robot pose. Secondly, in the case of there has more foothold points, select the points that have the greater margin stability and flexibility. Generate the foot end trajectory planning algorithm of the swing phase of the robot based on the obstacle convex hull algorithm. In order to ensure the stability and flexibility of the robot in next step, the robot pose adjustment algorithm is proposed. By adjusting the position of the robot’s gravity center and the ZMP point of the foot end to the maximal inscribed circle of the support polygon of robot truck to achieve pose adjustment in the flat and rugged terrain.In order to complete the comprehensive experimental study of the six-legged robot traverse in rugged terrain, a six-legged robot control program is designed and implemented. Through three sets of experiments to verify the effectiveness of the proposed method and theory for six-legged robot autonomous walking in unknown environment. The first experiment is implemented on the flat terrain, including projections and depressions obstacles. Through analysis of the environmental sensing time, sensing precision and posture change of the robot, verified the robot motion performance in such terrain. The second experiment is implemented on the columnar terrain, mainly verify the effectiveness of the foothold select algorithm and pose adjustment algorithm. The third experiment is implemented on the imitation hilly terrain, validating the various stages of a whole system comprehensivly. By stereo matching algorithm to recover the terrain point cloud information, based on triangulation algorithm to reconstruct the environment map of robot motion. By analyze the robot posture changes in the process of its moving, validate the reliability and validity of the posture adjustment algorithm. By comparing the experiment results, we found that the complexity of the scene impacts the efficiency of the foothold selection and trajectory planning of the robot, thereby affecting the traveling speed of the robot. Comprehensive experiments of the six-legged robot systems can fully verify the effectiveness of the stereo matching algorithm, three-dimensional reconstruction algorithm, foothold selection algorithm, trajectory planning algorithm and the pose adjustment algorithm which proposed in this paper.