BROWSE
Detail
Conference Paper
A Cyber-Physical System (CPS) Supporting Large-Scale Satellite-Drone Hybrid Application Development
Cited 0 time in 
Share 
Share
- Authors
- Soojeon Lee, Jeonggi Yang, Uihwan Choi, In Jun Kim, Yoola Hwang, Byoung-Sun Lee
- Issue Date
- 2024-10
- Citation
- International Astronautical Congress (IAC) 2024, pp.1-5
- Language
- English
- Type
- Conference Paper
- Abstract
- The adoption of machine learning methodologies, including deep learning and reinforcement learning, has expanded across numerous fields due to the progression in deep neural network investigations. For effective learning, the procurement of data on a large scale and its fast processing are essential. Nonetheless, in specific real-world scenarios like autonomous drone flight, training physical drones with an underdeveloped AI poses risks, is economically inefficient, and demands considerable time, thereby hindering the attainment of effective learning. To mitigate these issues, approaches based on simulation can be explored. However, simulations also have limitations in reflecting the diverse and unpredictable aspects of the real world. This paper proposes a Cyber-Physical System (CPS) that integrates the advantages of both the real and virtual worlds, ultimately enabling the coexistence and interaction of large-scale real/virtual objects especially drones and satellites. Our CPS architecture comprises three different types of entities: 1) real objects that exist only in the real world, 2) virtual objects that exist only in the virtual space, and 3) avatar objects that are functionally virtual but reflect the status of real objects. To facilitate the connection between real objects and their avatar counterparts, a CPS gateway is implemented. Our CPS is based on the Unreal Engine to simulate high visual realism and physical interactions among objects. However, the Unreal Engine operates on a substantial amount of compute resources, making it challenging to support the simultaneous coexistence of a large number of objects. To resolve this scalability issue, we propose a distributed CPS architecture that can operate on a computer cluster rather than a single computer. Utilizing our CPS, various RD activities and the development of diverse digital twin applications targeting large-scale drones and satellites, can be conducted more efficiently.
- KSP Keywords
- CPS architecture, Deep neural network(DNN), Digital Twin, Network investigations, Physical interaction, Real-world, Reinforcement learning(RL), Unreal Engine, Virtual worlds, application development, computer cluster