Reinforcement learning-based simulation and automation for tower crane 3D lift planning

Tower crane lift planning is important to timely provide resources to workplaces. However, previous planning approaches are still impractical because the lifting time of a plan is barely considered and the lifting path is frequently non-executable by operators. This paper describes a reinforcement l...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Automation in construction Ročník 144; s. 104620
Hlavní autoři: Cho, SungHwan, Han, SangUk
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier B.V 01.12.2022
Témata:
3D dynamic simulation
Motion planning
Path planning
Reinforcement learning algorithms: on-policy and off-policy
Reward function
Tower crane
Tower crane
Motion planning
3D dynamic simulation
Path planning
Reward function
Reinforcement learning algorithms: on-policy and off-policy
ISSN:0926-5805, 1872-7891
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Tower crane lift planning is important to timely provide resources to workplaces. However, previous planning approaches are still impractical because the lifting time of a plan is barely considered and the lifting path is frequently non-executable by operators. This paper describes a reinforcement learning-based method that incorporates the actuator system of a tower crane into spatio-temporal lift planning in three-dimensional virtual environments wherein various strategies of algorithm types and learning rules are tested. The results show stable and practical lift planning with a failure ratio of 3%, coordination ratio of 28%, and positive evaluation of lifting procedures by expert operators. In addition, the estimated lifting time shows a correlation of 0.6857 with the actual time from field observation. Thus, the proposed approach is promising for planning feasible lifting paths and estimating reasonable lifting times, which help generate and review lifting plans given the site conditions. •Reinforcement learning (RL) is proposed for automated lift planning of tower cranes.•Mechanical and operational properties of lifting are formulated for a crane agent.•Six modeling strategies are tested to train the RL agent in a virtual environment.•Executable lift planning is achieved along with reasonable estimations of task time.•RL-based simulation may help make and review lifting plans given site conditions.
ISSN:0926-5805
1872-7891
DOI:10.1016/j.autcon.2022.104620