*3.1. Cloud in the Loop Simulation (CILS):*

The ability to simulate a wide range of heterogeneous personal aerial vehicles (PAV) in the same virtual environment is critical and required to verify a variety of AI control algorithms even before their practical implementation on physical twin vehicles in digital twin infrastructures of future urban air mobility (UAM). One may deploy the AI control algorithms on actual vehicles and train them through practical flight testing in actual surroundings in order to improve the precision and calibre of neural network-based AI control algorithms (for example, neural Lyapunov control [18], deep reinforcement learning [19]). However, it frequently takes a lot of work and a considerable amount of time to collect enough flight test data for developing a competent AI control model for autonomous PAVs. As a result, one of the popular approaches is to develop an operational digital twin system for UAM (abbreviated as UAM-ODT) to replicate the actions of UAM vehicles in real-world settings within a shared virtual environment [20].

Inspired by the idea, we propose to adopt cloud-based solutions to develop a UAM-ODT system for a specific eVTOL PAV, called eVTOL KADA-UAM vehicle, under development by Konkuk Aerospace Design-Airworthiness Research Institute (KADA), Konkuk University, Seoul, Republic of Korea as shown in Figure 2 and its virtual environment of UAM-ODT as shown in Figure 3.

**Figure 2.** A digital replica of UAM vehicle in UAM-ODT infrastructure.

(**a**) In Flight

(**b**) Landing

(**c**) Vertiport

**Figure 3.** A visualization of UAM-ODT infrastructure. (The figures are excerpts from a video at https: //blog.naver.com/yy8661 provided by Hyeon Jun Lee, Konkuk Aerospace Design-Trustworthiness Institute, Konkuk University, Seoul, Republic of Korea (rain9138@gmail.com)).

The proposal is the overall cloud-in-the-loop simulation (CILS) framework that can simulate the operations of a multitude of heterogeneous UAM vehicles with completely different aerodynamics in a UAM-ODT system, and thus can be used for verification and training of AI control algorithms in virtual world before practical implementation. The overall conceptual CILS architecture is designed as in Figure 4. To simulate multi-mode operations of heterogeneous environment which consists of multi-vehicles with different dynamics and configuration, we adopted the virtualization concept in cloud computing paradigm to separate a multitude of SILS processes onto different VMs. A single SILS process encompasses a PX4-based autopilot multi-mode AI control module (abbreviated as PX4) and a JSBsim based aerodynamic module called Konkuk Flight Simulation-Digital Twin Dynamics module (KFS-DT). The encapsulation of these two modules is called a dynamics-control SILS package, which is deployed on a multitude of VMs. The KFS-DT module guarantees the concept of digital twin framework for K-UAM vehicles in which it captures a high-fidelity CFD dynamics model of each physical vehicle. The autopilot control PX4 module transfer controls of each vehicle **u** to the dynamic module KFS-DT. The KFS-DT module computes new vehicle states **s** and returns them to the PX4 module.

The PX4 updates the current states **s** to a VM controller module. On the other hand, the PX4 receives updated sensor data from the VM controller module to generate new controls **u**. On each VM, there is a VM controller module to handle the data transactions between the VMs with a physical server for operational control management and for the visualization of the virtual environment (called environment control center). The VM controller module in each VM transmits vehicle states s receiving from a control PX4 module in the same VM and receives sensor data or mission data to/from the visualization center using Airsim [21] for AI application and Unity™for visualization. While an environment controller module in the control center is designed to handle the operations of all vehicles in the simulated environment upon the ground control module, AI module, Airsim server and Unity client module. The scalability of this cloud-based simulation framework is guaranteed by an auto-provisioning cloud system.

**Figure 4.** Cloud in the loop simulation framework.

In this work, our focus is on the digital twin version of an eVTOL vehicle to capture flight operations in an urban air mobility for further studies on performance modeling and simulation of vehicle, fleet management and maintenance, traffic management and control, and emergency response and rescue. Thus, the detailed design of the overall cloud in the loop simulation framework running on a private cloud platform is presented in a comprehensive manner while the detailed design of the eVTOL vehicle in consideration regarding circuit and IT problem in the individual engine control systems is out of scope of the study. We consider how the dynamics of the vehicle and its corresponding control are simulated in a virtualized environment of urban air mobility to mimic the real-world flight operations for air traffic managements in urban areas.

### *3.2. Cloud Provisioning Hardware System:*

The hardware infrastructure of CILS is designed as shown in Figure 5, consists of main two components: one is virtual cluster (VC) disk image provisioning, and the other is auto provisioning virtual instance creator. Virtual machines existing in the same VC subgroup can be used by sharing the virtual disk image with homogeneous S/W as readonly. The numerous existing cloud systems provides GUI where users can build instances. It seems that this function simplifies the manipulation of creating instances on the cloud

system whereas it just repeats useless operation to prepare requisites and build VMs. The auto provisioning virtual instance creator based on infrastructure as a code (IaC) provides a consistent CLI workflow to manage hundreds of cloud services and customize simulation environments.

**Figure 5.** Cloud provisioning hardware system architecture.

Figure 6 shows the provisioning technology for virtual cluster images. The VC disk image provisioning based on union mounting technique can integrates one instance with diversified simulation virtual disk images depending on the simulation requirement such as PX4-Autopilot, JSBSim, Airsim, FlightGear and so on. Due to the orchestration of the cloud management, it can implement the communication via layer 2 or layer 3 between instances. Thanks to the capabilities of cloud provisioning and orchestration as designed in Figures 5 and 6, the UAM management can be maintained and stored on CILS Cloud Compute System complying the CILS framework in Figure 3.

**Figure 6.** Virtual cluster image provisioning technology.
