Study On The Kinect Based Grasping Path Planning For Service Robot

With the improvement of people's living standard and the aging of the population, research on service robot is becoming an important direction in robot area. Due to the complex and unstruc?tured problems of the home environment, the path planning for grasping faces new challenges.With the three-dimensional information provided by Kinect depth camera, a path planning system is designed for Baxter robot which enables it to grasp obeject in home environment.Firstly, the basic principle of robot operating system (ROS) is introduced, then the simulation and experiment system of Baxter robot is built based on ROS. For the visual sensor Kinect, the working principle of Kinect is studied, the visual coordinate system of Kinect is established and the Kinect parameters are calibrated. By taking Baxter's left arm as experimental object, the kinetic dynamics is established using the DH coordinate method. And the Newton-Raphson method based on singular robustness inverse is used to solve the inverse kinematics.Secondly, since the original depth information obtained by Kinect has many defects such as noise, cavity and object edge overflow, an improved depth image restoration algorithm is proposed on the basis of comparing the commonly used depth image filtering methods. In order to solve the error information in edge, a method based on Canny edge detection and morphological filtering is proposed to remove the error edge information. A joint bilaterial filter with depth threshold is used to remove noise value in the reliable area. Under the guidance of color image boundary, the iterative method is used to restore invalid depth points(holes and wrong edges), and it makes the three-dimensional(3D) information more quality and precise. Then, the Octomap is generated for robot path planning based on improved 3D information.Thirdly, methods of lazy collision check, modified nearest point selection, path optimization and intelligent sampling are introduced to improve the performance of the orginal bi-RRT* algo-rithm. The lazy collision detection strategy is applied to the nearest node and the optimal parent node selection process, which effectively reduces the number of collision check. The mod:ified nearest point selection makes full use of existing planning information, and performs adaptive trade-off between node expansion and path optimization. The path-optimized nodes are used as biased points in intelligent sampling, which makes sampled nodes closer to obstacle configuration,thus further reducing the path cost. Experiments show that the IB-RRTa* algorithm can converge to the optimal path faster while maintaining the high planning success rate.Finally, with the 3D spatial information provided by Kinect, the IB-RRT* algorithm is used to plan obstacle avoidance grasping path in simulation and experiment. The experimental results verify the effectiveness of the proposed algorithm and realize the robot avoids obstacles in the actual scene and completes the grasping task.