인간-AGV 충돌 위험을 고려한 강화학습 기반의 출하대기장 내 AGV 운영 최적화

초록

Automated guided vehicles (AGV) are a crucial component to achieve automation in logistics. AGVs can autonomously transport heavy pallets with the pre-designed rules, hence reducing the need for human resources in the logistics area. However, since AGVs cannot completely replace human workers doing maintenance tasks, inspections, etc., AGVs sometimes share operating areas with human workers. As such an example, we optimize the operation of an AGV in the dispatch area which has a human operator for final inspections. While most previous studies focus on efficient operation of AGVs, this study considers the possibility of human-AGV collisions as well as efficient operation. We also propose a PPO-R algorithm to prevent conservative behaviors of AGV when introducing a collision penalty. Numerical experiments show that PPO-R can maintain throughput while reducing the number of potential collisions.

키워드