| Title |
Reinforcement Learning-based Excavation Trajectory Generation Algorithm Utilizing Soil Data |
| DOI |
https://doi.org/10.5573/ieie.2025.62.2.85 |
| Keywords |
Autonomous excavator; Excavator trajectory; Task planning; Reinforcement learning(RL) |
| Abstract |
This paper presents a novel reinforcement learning (RL)-based control framework for excavator trajectory planning, incorporating dynamic soil condition data. The primary challenge in the proposed algorithm lies in addressing soil state dynamics, which pose significant difficulties in achieving stabilized convergence during deep reinforcement learning training. To overcome this issue, the interaction between the excavator and the soil is modeled using the fundamental equation of earthmoving (FEE). The RL-based trajectory control model is trained using randomly generated soil parameters, enabling autonomous excavation task planning under varying soil conditions. The performance evaluation results demonstrate that the proposed method effectively adapts to diverse soil characteristics, thereby enhancing its applicability to real-world excavation operations. |