Research And Implementation Of Indoor Autonomous Mobile Grabbing Robot Based On ROS

As an interdisciplinary research of application,the mobile grabbing robot has broad application prospects in logistic,industry,smart agriculture,household service and other industries.Therefore,it is of great significance to carry out relevant research on mobile grabbing robot and implement a mobile grabbing robot system for improving the intelligence level of robots in our country and the performance of deep learning algorithms.However,in the process of actual research and application,there are some certain limitations in some of the current mainstream algorithms.For example,When building the map of complex environment,due to the limited field of lidar’s view,there is a large dead zone of detection in the scanning range of lidar,which cannot effectively detect the low obstacles.Furthermore,there are different degrees of drift in the constructed environment map.In terms of 6D object pose estimation,since most robots in real scenarios are implemented based on embedded platform,which has a limited computing ability and cannot meet the hardware requirements of algorithms,so there are some difficulties in the application of algorithms.In order to solve the above two problems,this thesis mainly does the following work:From the perspective of multi-sensor fusion,the laser-based simultaneous localization and mapping(SLAM)algorithm based on RGB-D Information enhancement is proposed.The algorithm constructs a fake lidar data sequence through the depth image,which is fused with the lidar data through the Bayesian estimation method to build the map,thus effectively detecting the low obstacles;The loop closure structure is constructed by visual features extracted from RGB images.Combined with the particle feature,the structure can effectively detect loop closures in the process of mapping,thus alleviating the drift phenomenon.In addition,two hyperparameters are introduced in the structure to improve the robustness of algorithm in different scenarios.From the perspective of model lightweight,a lightweight 6D pose estimation algorithm called MobileNet 6D is proposed.In the design of the network structure,the DSCENet(Depthwise Separable Convolution Expansion Network)is proposed to infer the 2D image coordinates of projected 3D bounding boxes of targets,followed by the perspective-n-point(PnP)algorithm to estimate the pose of targets.On the FAT public dataset,the effectiveness of the algorithm proposed in this thesis is verified by comparative experiments,that is,the inference speed of the model is improved while ensuring the prediction accuracy.Based on the hardware resources of the laboratory,the mobile grabbing robot system is designed and implemented.The algorithms proposed in this thesis are encapsulated through ROS,and the corresponding subscription and publishing models of topics are implemented.The move＿base framework is used to realize the robot’s autonomous movement and navigation,and the Moveit! development platform is used to implement motion planning of manipulator,with the electric gripper to grasp the target.Finally,the performance of mobile grabbing robot system is tested in the office environment.