Autonomous Navigation Of Mobile Manipulation Robot In Multi-floor Building With Elevator

With the development of urbanization in China,most of the people live in city spend most of their time inside buildings,which create huge application requirements for service robots.For most of these application in buildings,the robots have to autonomous navigation on multiple floors.This dissertation designs an autonomous navigation system for mobile manipulation robot,researches the environment states detection and the elevator button detection and operation technology,makes the robot perform a wide range navigation inside the building by taking elevator itself without modifying elevator devices.The navigation system is divided into five parts,including task planner,chassis navigation,manipulation,environment states detection and robot hardware.All the function modules in the system are controlled by a state machine.Chassis navigation is modified from ROS(Robot Operation System)navigation stack.The maps of multi-floor are aligned by choosing the feature points in different floor and publish by multiplex mechanism,solved the robot transition problem between maps.Manipulation charges elevator button detection and operation.The proposed button detection algorithm based on stroke width transform for regions extraction,then these regions recognized by convolutional neural network trained by standard font library character.For operation,this dissertation proposes algorithm plan the path and pose from the 3D point cloud of target button with random sample consensus.The experiment shown the proposed algorithm can detect and operate the unknown elevator panel button.Environment states detection including person motion for safety,elevator door state and floor state for stable navigation.For person detection,based on one stage detection model YOLOv3 Tiny,by replacing the backbone with Mobile Net V2 and changing the loss function with state of the art function,succeed to compress the model with advancing the detection performance.For elevator state detection,this dissertation proposes the door state detection algorithm with laser scanner and floor state detection algorithm with the barometer,performs accurate detection in experiment.Finally,this dissertation completes the hardware selection and assembly design of robot,deploys and verified the system by ROS on it.The experiment shown the navigation system proposed by this dissertation can make the mobile manipulation robot navigate independently between floors inside the building with elevator.