'Synthetic Data' Teaching Robots About Reality

Existing data included information about past events or real-world data that robots and AI learned directly. In contrast, synthetic data is generated in a virtual environment to prepare for unpredictable future situations, contributing to preventing accidents in autonomous vehicles and improving efficiency in manufacturing processes.

This allows robots to learn various scenarios they have not experienced before, reducing the likelihood of errors in real environments. In particular, it plays a crucial role in increasing the adaptability of robots in complex and unpredictable situations, thereby boosting industrial productivity. As such, synthetic data has become an essential element for expanding the scope of AI learning and solving real-world problems.

For example, synthetic data can virtually reproduce rare accident situations that may occur during autonomous driving, or train robots to pre-recognize subtle component deformations that may occur on a manufacturing line. This forms the basis for robots to react instantly and accurately in real environments.

For example, if an autonomous vehicle needs to learn about a rare situation like a suddenly overturned truck, it would require setting up the truck on a real road, controlling traffic, and deploying specialized personnel. Such a process would demand immense cost and time, making it practically difficult to implement.

Synthetic data solves these problems by endlessly generating diverse weather conditions and accident scenarios in a simulation environment. In particular, reducing the 'domain gap'—the difference between real and virtual data—to 0% is emerging as a key challenge in the industry. The goal is to ensure that robots trained in virtual environments operate without errors in real environments.

In manufacturing processes, various variables such as component contamination, scratches, and changes in lighting environment can occur. However, robots often fail to recognize these subtle changes, leading to process shutdowns or 'jam' phenomena.

This limitation in robot perception can be overcome using synthetic data. Synthetic data trains robots with all possible variables of a component in advance, enabling them to accurately recognize the component and perform the process in any situation. This is a crucial key to significantly improving manufacturing productivity.

Once this gap is completely resolved, robots trained in virtual environments will operate perfectly in real environments. This lays the foundation for robots to perform complex tasks without human intervention.

Major global companies such as Fanuc, Microsoft, and Boston Dynamics are investing heavily in domain gap resolution technology. This technology plays a decisive role in maximizing robot utilization and innovating productivity in various industrial fields, including autonomous driving and smart factories.

Skyworldwide, as a listed company, is responsible for the entire physical AI architecture, while its affiliate Skyintelligence utilizes NVIDIA Omniverse to monetize 3D digital twin data. Based on this data, Skyworldwide collaborates with industrial robot company ABB to build the overall physical AI process.

This division of roles among affiliates efficiently integrates the entire process from data generation to operation and actual robot application. Skyworldwide focuses on core technology development and system integration, while Skyintelligence adds expertise in data monetization using NVIDIA's technology stack.

Skyworldwide focuses on improving productivity in industrial settings and actively uses synthetic data to optimize robot assembly processes. For robots to perform complex assembly tasks perfectly, they need to generate diverse data—beyond simple physical laws—such as material, lighting, and position, in a virtual environment.

When humans provide basic guidelines, AI uses them to create countless scenarios, establishing an environment where robots can operate stably in any situation. This prevents robot malfunctions and accidents due to a lack of data, contributing to a significant increase in manufacturing productivity.

Data scanned by robots is implemented in a virtual environment with real-like textures, structures, and lighting. This technology was recognized in the global market by winning the Grand Prix at 'Viva Technology 2024'.

Originally used for advertising production, this technology is now expanding its application to manufacturing data. Since robots scan with extremely high precision, even minute details of real objects can be perfectly reproduced in the virtual environment, playing a crucial role in building training data for industrial robots.

The first stage is data storage utilizing infrastructure (GPU), followed by efficient operation, systematic management, and thorough security. The final stage is the intellectualization and structuring process, which transforms data from mere information into 'knowledge'.

This can be compared to building a house. The foundational stages of land and framework represent the infrastructure, followed by storage, organization, and security. Finally, it reaches the intellectualization stage where AI analyzes data to derive optimal answers. This architecture aims to be a platform that handles not only robot data from manufacturing plants but also all data concerning companies, products, and humans.

In particular, this structure manages the entire lifecycle from data generation to utilization, focusing on helping companies create substantial value beyond simply storing data. It maximizes the potential of data through organic linkage at each stage.

While traditional relational databases (RDBs) organize data in rows and columns like an Excel spreadsheet, graph DBs connect data with nodes and edges to identify hidden relationships.

This method helps in optimal decision-making by analyzing the context between data that could not be understood from simple tabular information. For example, in the financial sector, it is used to find complex relationships like money laundering, or to instruct industrial robots on the most efficient tasks.

To prevent such errors, Skyworldwide does not allow individuals to input personal opinions or thoughts directly.

Instead, it adheres to a method of deriving relationships between data using only accurate and quantifiable data. This ensures objectivity and reliability in analysis, focusing on enhancing the accuracy of causal relationship analysis.

Existing rule-based detection methods were vulnerable to fraudsters who split transfers into multiple accounts, sending 9.99 million won each time. However, graph DB visually connects and identifies the relationship where the same amount of money converges into one account from multiple accounts at the same time.

Cases have shown that such network analysis can effectively prevent damage. It is also used in airport customs to identify repeat offenders through同行者 analysis, demonstrating excellent ability to analyze and predict complex crime patterns.

Graph DB goes beyond simply looking at individual data; it finds hidden connections between data to detect anomalies occurring within complex relationships at an early stage. This becomes an innovative tool for preventing and investigating not only financial fraud but also various forms of organized crime.

In particular, it shows strength in responding to sophisticated crime methods that were difficult to identify with existing systems. This significantly contributes to strengthening social safety nets and reducing public harm.

This is their exclusive competitive advantage, with the core goal of securing a competitive edge within an integrated system.

By adopting Microsoft Azure technology, for example, they secure technical capabilities that meet global standards, thereby validating the system's stability and scalability. This strategy demonstrates an intention to target the market by providing a complete solution rather than competing on individual technologies.

Such an integrated architecture encompasses the entire data lifecycle, providing customized solutions optimized for clients. While global companies focus on individual technological elements, Skyworldwide plans to expand its market share by leveraging the efficiency and flexibility of its overall system.

This strategy involves making their technology publicly available for anyone to use, aiming to be judged solely on their technological prowess and lead industry standards.

Similar to Elon Musk's open-source strategy, it intends to prove their superiority and preempt the market through technological diffusion. This aims to enhance the status of Korean technology and strengthen its competitiveness in the global market.

This is to meet the specialized environment and security requirements of each factory.

Skyworldwide plans to re-export its globally recognized technology in the form of customized data management platforms to Korean factories. Through this, they aim to support the smartification of domestic manufacturing and explore expansion into the global market.

Even with a physical body (robot), it is useless without the experience (data) to perform tasks in a real factory. In this area of creating data-driven experience, a Korean company, Skyworldwide, has seized an opportunity.

A model of collaboration across technological domains, like Google and Boston Dynamics, is spreading in the global market. Within this trend, Skyworldwide builds partnerships that provide data solutions to robot hardware companies, creating synergy.

This emphasizes the importance of a collaborative model where different strengths are combined to create new value in a complex industrial environment where a single company finds it difficult to cover all technological domains. Skyworldwide's strategy is to establish itself as a key player in the global robot market through this collaborative model.

While simply utilizing open data is easy, the key is to securely manage and leverage fully confidential data.

Companies need solutions that meet four core requirements: operations, security, on-premise environment setup, and data generation. Skyworldwide addresses these needs by securely monetizing hidden confidential data, thereby contributing to maximizing enterprise productivity and efficiency.