RACK-AWARE, PAYLOAD-AWARE LOCAL PATH PLANNING ALGORITHM FOR WAREHOUSE ROBOTS IN MULTI-ROBOT, HUMAN-SHARED ENVIRONMENTS

The paper addresses local path planning for autonomous mobile robots that carry loads in a warehouse shared by people and several robots. A planar LiDAR mounted close to the floor observes only the legs of a load-carrying rack, so the costmap marks the space beneath the rack as free and a loaded robot may strike the deck. We propose the RAP method (Rack-Aware, Payload-aware Planning): a two-channel convolutional neural network separates rack legs from people and other robots, the rack pose is recovered from two to four legs by a Procrustes estimator, confirmed with a depth camera, a 3D LiDAR or a monocular camera, and projected into the costmap as a load-carrying volume with a certified safety margin. Depending on the payload height, the robot either passes under the rack or goes around it. The estimator is proved to attain the Cramér–Rao bound, and all theoretical results are verified numerically. In 640 paired episodes RAP achieved 95.0 % success with 1.9 % safety violations, and in Gazebo/ROS 2 experiments the loaded robot never struck a rack deck.

Hali tarjima qilinmagan