Automation of crane control for block lifting based on deep reinforcement learning

Abstract

In shipyards, blocks are controlled by connecting the crane and block with wires during block erection. During block lifting, if a block is not carefully controlled, it will cause damage. Block lifting using crane operation is performed by controlling the number of wires, hooks, and equalizers. Consequently, predicting stable block lifting is difficult. In this study, we proposed a control method to determine static equilibrium. Initially, an algorithm for finding the initial equilibrium block state (IES algorithm) was proposed, followed by deep reinforcement learning (DRL) based method for block lifting. The position, orientation, angular velocity of the block, and hoisting speed of the wires were applied as the DRL state. The control input of the crane was calculated by deriving the hoisting speed of the wires. To verify the proposed method, comparative studies on the application of the IES algorithm were carried out, and further block movement was compared. Conclusively, the proposed method effectively increased block lifting safety.

Policy Optimization for Waste Crane Automation From Human Preferences

“…There has also been extensive research on automation for various types of cranes [15], including tower [16], gantry [17], [18], and overhead cranes [19]. Chun et al introduced a method that integrates deep reinforcement learning with an algorithm for identifying static initial equilibrium states to automate the lifting of large blocks by cranes, a typical operatordependent task [20]. Cho et al described their strategy for automating tower-crane-lifting operations and estimating the lifting times at construction sites.…”