| Title |
Reinforcement Learning-based Local Task Planning Algorithm for Excavation Automation |
| Authors |
신민규(Mingyu Shin) ; 조준형(Junhyung Cho) ; 정소이(Soyi Jung) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.1.105 |
| Keywords |
Excavator; Automation; Excavation planning; Reinforcement learning |
| Abstract |
Excavation work plays a crucial role in various fields, making it essential to ensure both efficiency and stability. In this paper, we propose a local task planning algorithm based on Proximal Policy Optimization (PPO) for leveling operations. This algorithm leverages reinforcement learning, generating and training on a grid map within a work area defined by the actual specifications of an excavator. The excavator acts as an agent, receiving terrain information from the grid map and determining the areas where digging is required and the amount of material to be excavated. The leveling operation consists of two tasks: digging and dumping. Independent models were trained for each task, and the models were then applied sequentially, starting with digging followed by dumping. Experimental results demonstrated that each model successfully completed the leveling operation by iterating tasks until the target height was achieved. This algorithm is expected to maximize work efficiency and minimize errors during leveling operations, making it applicable to various working environments. |