http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
유전 알고리듬과 교환 휴리스틱 알고리듬을 이용한 무인이동로봇의 잡샵 생산 일정 계획
왕준경(Jun-Kyung Wang),어규호(Gyuho Eoh),박태형(Tae-Hyoung Park) 제어로봇시스템학회 2022 제어·로봇·시스템학회 논문지 Vol.28 No.2
Solving a job shop scheduling problem (JSSP) entails allocating entire jobs to machines to minimize the processing time. Generally, the transit time between machines is not considered in the conventional JSSP because it is shorter than the processing time. In a real production environment, however, the transit time of an autonomous ground vehicle (AGV) is long due to the need to solve random problems, such as finding an indirect route or avoiding collisions. Moreover, AGVs sometimes need to transport the job on directional roads with upstream and downstream constraints. Therefore, we present an AGV job scheduling algorithm considering the transit time. This study considers not only the transportation times of the jobs from one machine to another but also collisions on limited roads, similar to in a real environment. A near-optimal job sequence was obtained by combining the genetic algorithm (GA) with simulation and was comparing this with the traditional method. Additionally, this paper presents heuristics that deal with deadlocks in the process of combining JSSP and simulation. The combination of GA and heuristics was tested using various simulations.
차선 변경 판단을 위한 증강 데이터 기반 강화학습 방법
김민성(Min-Seong Kim),어규호(Gyuho Eoh),박태형(Tae-Hyoung Park) 제어로봇시스템학회 2021 제어·로봇·시스템학회 논문지 Vol.27 No.8
In a lane change decision-making technique based on RL (Reinforcement Learning), most data is related to keeping a lane; on the contrary, the data related to lane change is sparse. Consequently, an ego-vehicle fails to learn its policy. To overcome this issue, we propose a new RL-based lane change decision-making technique that uses a safety inspection module and augmented data. The safety inspection module restrains the collision of the ego-vehicle and generates virtual augmented data acquired from lane change situations. The introduction of virtual augmented data mitigates sparse lane-change data problems, which results in performance improvement. We simulated the proposed technique using an open-source simulator called CARLA.