Tristan Chatelin, Eletroteknik

​Title: Grasping with a mobile manipulator using 3D vision

Examinator: ​Yasemin Bekiroglu

Robotic grasping and manipulation is a well-studied topic, but often it’s implemented using stationary robotic arms. Using a mobile manipulator for grasping and manipulation enables significant increase in the capabilities of the robot. This thesis tackles the challenging task of grasping a given object with a mobile manipulator. Our approach maintains a good trade-off between the arm motion, base motion and execution time. This work is done using the data from a stereo camera which oversees the scene. 

Firstly, an instance segmentation algorithm is used to recognise the objects which are available in the scene. Next, one object among the batch of detected objects is selected. Then, the mask of the selected object is used to generate a point cloud which only contains the points which belong to the object. This point cloud is then used as input for a pose estimation algorithm to find the 6D pose of the object. Then, a set of grasp poses is generated based on the detected object pose.

The most appropriate grasp pose is chosen from the set of grasp pose knowing the actual pose of the gripper based on a defined selection criteria. The motion of all the joints of the robot is computed to follow an end-effector trajectory which is a straight line between the initial pose of the gripper and the selected pose. Our real robot experiments lead to successful grasp execution results with high grasp quality and a low computation time.

Tristan and Yasemin
Password: 1234
Kategori Studentarbete
Plats: Online
Tid: 2022-10-06 13:00
Sluttid: 2022-10-06 14:00

Sidansvarig Publicerad: må 03 okt 2022.