UC Riverside

Cargo transloading is an important part of the transportation, and the increasing needs of transportation is a challenge to the transloading capacity. Therefore, automating the process to reduce the manpower consumption and improve the efficiency of transloading is of interest. Cooperative transloading is a common method for bulk cargo transloading, which refers to a robot inside the container that transfers the bulk cargo to a convenient position that the crane can reach. Cooperation improves the crane grabbing efficiency, and avoids the bulk cargo being in the container areas that the crane cannot reach. In this dissertation, a method is proposed to recognize and locate the container hatch from point cloud, to unify the reference frame of the crane and robot, to determine the initial robot pose in that shared frame, and to localize the robot as it maneuvers in the container in real-time during the transloading. The main contributions of this research fit into two categories.

Frame Definition and Initial Pose Determination: To establish a reference frame recognizable by both the crane and the robot, a portion of this dissertation focus on extracting the hatch from the robot point cloud. One hatch corner and the hatch edges define the origin and axis of the shared working frame for the robot and crane. To find the hatch, the 3D point cloud scanned by the robot is rasterized into 2D data, preserving the relative position information of the hatch. A method based on the Hough Transform is used to determine the initial point cloud translation and rotation with respect to the hatch using the 2D data. With the determined translation and rotation, the common reference frame is defined, the point cloud is re-coordinatized into this frame and the initial robot pose can be determined and expressed in this frame.

Real-Time Localization: The transloading starts after determining the robot initial pose. The robot moves in the container. To avoid collisions, the robot real-time position needs to be reported to the crane. In this dissertation, a basemap is created initially using the point cloud scanned for determining the initial pose. Later, when the robot moves, its LiDAR scans are matched with the basemap using an ICP algorithm to determine the robot pose in real-time. To achieve more reliable matching results with the ICP algorithm, several approaches are compared for roughly aligning the real-time scans to the basemap before using ICP algorithm, including the use of LiDAR-estimated poses and velocities, and the use of IMU measurements to calculate the pose change. The experimental comparisons of those methods are assessed to determine the most suitable one for cooperative transloading.

In addition to the analysis and development of new methods for hatch recognition, cooperative frame definition, and real-time localization in a non-stationary frame, this research has developed a fully functional real-time prototype implementation.