Data-Driven Surface Classification for Differential Drive Autonomous Guided Vehicles

Abstract

As a result of the digital transformation, the degree of automation in production environments is constantly increasing. The automation of logistics processes offers great potential for optimizing material flows within production. However, there are strong requirements regarding the reliability of automated guided vehicles (AGVs). This paper investigates if it is possible to identify the surface of an AGV based on control and measurement parameters compared to target parameters. The vehicle concept is based on differential drive and is used for the autonomous transport of goods within and between production halls. The information about the surface can be used to ensure a better and safer interaction with the environment. The creation of a dataset consisting of a low and a high friction surface is described. Furthermore, the pre-processing of the data which covers the time window length, resampling and data scaling is described. Finally, different algorithms for multivariable time series classification are benchmarked on the created dataset. Experiments show that the state of the art algorithm HIVE-COTE 2.0 provides the highest accuracy on the test dataset with 97.5%. Finally, a validation is performed by reviewing sequences with low confidence levels.