Keywords

1 Introduction

The manufacturing industry is undergoing a digital transformation process where different technologies are converging towards flexible, individualized, and efficient production. For this purpose, the digitization of assets in industrial processes is essential in the creation of a collaborative environment to generate greater added value. Under this paradigm, the concept of Industry 4.0 is born where the integration of heterogeneous technologies is a requirement to achieve more efficient production [1]. In this sense, Cyber-Physical System (CPS) are key to merge the physical level with the digital level enabling the interconnection of the entire value chain [2]. The Industrial Internet of Things (IIoT) concept is being developed where policies and methods for data and information transmission are used to enable a collaborative manufacturing.

Within this context of digitization and connectivity, industrial handling is an asset where coordination of other manufacturing processes is needed in order to be efficient. To achieve efficient transportation, data generation of the different processes through SCF is necessary to obtain information for decision making. In this sense, decision making can be executed by an expert system because the necessary information is digitized. However, for full automation of internal transportation, autonomous vehicles capable of safely navigating the manufacturing plant are required. These vehicles are referred to as Autonomous Mobile Robots (AMR) in the scientific literature. AMRs are equipped with different sensors that allow them to recognize the environment as well as some on-board intelligence to navigate safely.

However, the adaptation of AMRs to industrial environments is unknown due to the heterogeneity of machine distributions as well as the dynamic location of personnel. Nevertheless, the evaluation of AMR behavior is essential for cost forecasting and production planning. In this context, Digital Twins (DT) enable the virtualization of an environment as well as the modeling of AMRs. By using DT, it is possible to analyze trajectories, times and even battery consumption, which are key to planning and optimizing the production activity.

In this context, there is a study [3] which analyzes the behavior of a single AMR in a DT compared to a real robot. In this study, two scenarios with static obstacles are recreated, reaching a high level of similarity; however, this work does not analyze the behavior with several AMR simultaneously. Moreover, the navigation of an AMR fleet presents several peculiarities in the field of navigation, planning or simulation. For this reason, this work studies the navigation behavior of an AMR fleet making use of a DT which is a novelty in the literature to the authors’ knowledge.

In this work, in addition to the creation of the DT with several AMRs, we intend to evaluate the behavior of two different local navigation algorithms. The objective of this study is to establish which one offers lower transport times once its parameters have been configured and adapted to avoid collisions between them. Furthermore, to make the evaluation comparable, it is performed in the same industrial simulation environment with identical case studies supposing the same requirements for AMRs navigation.

2 Case Study

AMRs are a fundamental asset for internal transportation in the industrial plant [4]. To perform this demonstrator and study with a fleet of AMRs, an industrial plant for the manufacturing of parts will be recreated virtually. For this purpose, the parts need to be moved from one machine to another to be stored in a warehouse.

With this case, the objective is to analyze the trajectories and times associated with the autonomous transport of these vehicles. However, in order to analyze the behavior of the navigation algorithms with moving obstacles, a scenario of small dimensions has been recreated to maximize the encounters with the AMRs.

2.1 Description of the AMR Used in This Paper

The autonomous industrial vehicle used is a non-holonomic mobility robot [5]. In this case, the robot has two independent driving wheels providing two Degrees Of Freedom (DOF) as opposed to the 3 DOF of the holonomic robots (i.e., two for position and one for orientation).

As for its kinematics, it is based on a differential traction movement by means of two conventional fixed wheels. However, at its ends, it has off-center steerable wheels for load distribution and stability, which are not driven.

The driving wheels are located on the sides of the geometric center of the robot, allowing it to rotate on itself without the need to move. In this way, a high mobility in confined spaces is achieved, which is considered in the navigation algorithms.

In addition, this autonomous vehicle under study is marketed by the company Mobile Industrial Robots (MIR) and is widely used in industry. Specifically, the model is the MIR100, which is capable of carrying 100 kg of payload with a maximum speed of 1.5 m/s.

2.2 Mission Definition

To evaluate both algorithms, the environment and objectives must be the same to be comparable. For this purpose, a mission has been established for each of the AMRs. In this way the environment and circumstances are identical, in order to attain comparable results for both algorithms.

For this case study, all AMRs are sent missions at the same time, moving to two different targets. The mission constraints stipulate that two vehicles cannot be parked on the same machine. The missions defined for each AMR are shown in Table 1.

Table 1. Definition of missions for the evaluation of navigation algorithms.
Full size table

Once the missions and the case study have been established, we proceed to the resolution of the problem.

3 Methodology

For the comparative analysis of the navigation algorithms, the ROS ecosystem has been used, where the interconnection between a set of libraries and tools has been established. The navigation stack is summarized in the move_base_node where it receives the positioning information according to the theoretical framework [6].

To integrate multiple robots, the authors have decided to create prefixes in order to load the same features and parameters without cloning the code. In this way with a single XML file, we can reproduce the AMRs in the same simulation environment.

The virtualized dynamic simulation environment is realized with gazebo where AMRs are spawned for navigation. The environment has been designed and exported in Spatial Data Files (SDF) for interpretation in the ROS ecosystem [7]. The manufacturing plant is quadrangular in shape with a size of 16 m on a side.

After the creation of the environment, the AMR poses have been defined at the points of interest defined in the mission.

4 Results and Discussion

This section shows the sailing times of each AMR, as well as the trajectories for comparison purposes.

4.1 Trajectories

The AMR trajectories is an indication of how the navigation algorithms work. In this case, the actual AMR trajectories are shown as a result. To reach the resulting trajectory, the local planning algorithms also rely on the global trajectory, as well as its nearest environment. The authors have employed the rviz tool to obtain the resulting trajectories as well as the mapped environment.

First, the trajectories with the DWA local planning algorithm will be shown. For the sake of clarity Fig. 1 shows only the resulting trajectory, ignoring the global trajectories of each AMR.

Fig. 1.
figure 1

AMR trajectories in the digital twin with the DWA local planning algorithm.

Full size image

According to Fig. 1, each AMR is differentiated with a number. Furthermore, the mission has been divided into two maps as a consequence of the two trajectories. In the same way as in the DWA algorithm, the trajectories of the local planning algorithm TEB are presented in Fig. 2.

Fig. 2.
figure 2

AMR trajectories in the digital twin with the TEB local planning algorithm.

Full size image

In addition, the maps in Figs. 1 and 2, which have been generated by rviz, show the cost map and its value by means of a color gradient in shades of blue. However, the dynamic obstacles caused by the AMRs have not been represented for clarity.

4.2 Navigation Times

In addition to the trajectories, the time taken to navigate the AGVs has been measured with the DWA and TEB algorithms, the results of which can be seen in Table 2.

Table 2. Comparison of navigation times between the DWA and TEB algorithms.
Full size table

4.3 Discussion

The TEB algorithm, being based on the temporal optimization of trajectories, seeks more optimal times compared to the DWA algorithm. Therefore, the results are consistent with the mathematical principles on which these two algorithms are based.

5 Conclusions

DTs are key for the virtualization of the environment and the generation of information for intelligent decision-making. In this sense, two navigation algorithms have been analyzed: DWA and TEB. For this purpose, an industrial environment has been virtualized with three AMRs in order to acquire the trajectories and navigation times for the same mission. Consequently, the ROS ecosystem has been used to evaluate the algorithms. According to the results obtained, the TEB algorithm offers reduced operation times than the DWA for dynamic obstacles. Therefore, it can be concluded that TEB is still better when dealing with moving obstacles. Also, the trajectories with TEB are more suitable for the objective function, although the fulfillment of the constraints is not guaranteed. In addition, a web interface has been developed to allow interaction between users and the DT within the ROS ecosystem, facilitating its use. Therefore, the initial objectives of this work have been successfully completed by analyzing and extracting interesting conclusions on local navigation algorithms for AMRs fleet navigation using an interactive digital twin.