Robot-to-Human Construction Tool Handover Grasp Prediction for 6-DOF Robotic Arm with Parallel Gripper

Workers often spend significant time locating tools on construction sites, resulting in productivity losses. With the continuous advancement of robotic technologies, robot-to-human handover offers a potential solution to address this issue. However, existing robot handover methods face challenges in handling the irregular shapes and dangerous nature of tools, as well as the dynamic and complex environments of construction sites. To tackle these challenges, this research proposes a method that enables robots to effectively predict and grasp construction tools while ensuring safety. A grasp dataset consisting of four commonly used construction tools-pliers, screwdrivers, wrenches, and safety glasses-was generated in a simulation environment. During the annotation process, grasping areas for robots and humans are distinguished to reduce potential risks in handover processes. The dataset includes multi-angle RGB-D images and detailed annotated grasp poses specifically designed for robotic grasps. This proposed dataset provides a foundation for training and validating grasp prediction models, enabling robots to handle tools with irregular shapes in various scenarios. The proposed approach integrates data annotation, scene generation, grasp pose prediction, and motion planning to achieve safe and efficient handovers. Validation experiments conducted in a simulation environment using a UR5 robotic arm equipped with a parallel gripper demonstrate a 92.5% success rate. This research provides a foundation for the application of robot-to-human handovers on construction sites and highlights the potential of using virtual datasets to address real-world challenges.