Swept path analysis is an analysis method that calculates the movements and paths of each part of the vehicle when the vehicle moves. Swept path analysis calculates the path that each wheel passes and the space that the vehicle can pass during the tur...
Swept path analysis is an analysis method that calculates the movements and paths of each part of the vehicle when the vehicle moves. Swept path analysis calculates the path that each wheel passes and the space that the vehicle can pass during the turn.
Oversized cargo has various considerations in land transportation. In particular, because the cargo is very large, it is important to identify transportation interferences due to the shape of the road, such as width and curvature of the road. If the width of the road is shorter than the width of the module or the length of the module is too long to cross the curve, the cargo must be transported by another route.
Existing swept path analysis software enables path analysis for oversized cargo transportations. However, although path analysis for vehicles such as trucks and trailers is possible, there is no analysis platform for SPMT, which is an omnidirectional vehicle commonly used for oversized cargo transportation. In addition, the current process of swept path analysis is very passive. Analyze the trajectories by drawing roads and manually placing the vehicle. In order to identify the feasibility of oversized cargo transportation on a specific curved road, it is necessary to manually check the trajectory by changing the layout of the vehicle.
Deep reinforcement learning is an area of machine learning. This is a way to utilize deep neural network structure in existing reinforcement learning algorithm. The agent recognizes the current state from the environment and selects the next action to maximize the reward by the deep neural network. Deep reinforcement learning is useful when states are expressed in higher dimensions. This method is also used in the field of simulation-based optimization. However, no deep reinforcement learning has been applied to swept path analysis.
In this study, we design a system that can analyze SPMT, an omnidirectional vehicle as well as a vehicle provided by the existing swept path analysis software. The system is also used to automate swept path analysis for oversized cargo transportation on an deep reinforcement learning basis. Ultimately, we develop an artificial intelligence system that directly identifies whether the vehicle can pass through the road. To verify the developed system, we compare the trajectories of the same vehicle with the existing swept path analysis software. Additionally, the ground truth is the result of identifying the feasibility of transportation through manual operation. Through this ground truth data, we verify the performance of the artificial intelligence system that identifies the feasibility of oversized cargo transportation. The purpose of this paper is to increase the efficiency of the process for identifying the feasibility of oversized cargo transportation.