Evolving now for future growth!
The AI Tech Team

The AI Tech Team is at the forefront of applying AI technology to automate and optimize operations throughout Hyundai Glovis.
The team outlined its key tasks and plans for 2023 to lead the company’s digital transformation in support of new businesses.

AI technology is leading the current changes in the logistics industry. At this critical juncture, when the use of AI-based robot technology is emerging as the key to building and managing the digital ecosystem and when the AI technological environment is rapidly evolving, Hyundai Glovis is preparing to apply AI technologies to all its business areas. The AI Tech Team, under the Future Innovation Technology Center, stands at the helm of this technological transition.

The AI Tech Team supports domestic & overseas CC, KD centers, machine vision, TM AI, auto-biz pricing, EV ESS, and maritime operations. In addition to these, the team also supports new businesses related to automotive parts and finished vehicles, logistics, maritime, auto businesses, hydrogen, and EVs, among other areas. By 2021, the team had already verified the effectiveness of its AI technology and in 2022 it began to lay the foundations for the practical and onsite application of AI. Based on these efforts, the team plans to expand its utilization of AI technology to all business divisions by the end of 2023.

Let’s learn more about the AI Tech Team’s achievements regarding domestic & overseas CC, KD centers, machine vision, and auto-biz pricing.

Image-based performance improvement and logistics automation
Machine vision

Machine vision is a technology that enables machines to perform perception and judgement-based tasks that are usually performed by humans. After extracting and classifying images or videos obtained from cameras attached to various pieces of equipment, robots and drones use deep learning to improve performance and automate related processes. The AI Tech Team plans to develop AI technology and apply it to logistics automation, using it to: minimize failure rates through the identification of defects in finished vehicles; automate inventory inspections by verifying box labels in logistics warehouses; improve the performance of existing QR and barcode recognition systems; and prevent safety accidents.

In order to perform machine vision tasks, it is necessary to have a deep understanding of AI technology and its implementation as well as logistics and the relevant needs out in the field. From a technical perspective, deep learning frameworks such as Python, TensorFlow, and Pytorch are mostly used for program development, and Convolution Neural Network (CNN) is used for image learning and processing.

The AI Tech Team is currently focusing on image classification and text recognition. Image classification technology is largely being used to identify defects in finished cars and to allow for vision recognition using CCTVs. For the exterior inspection of finished cars, the team is preparing to implement Proof of Concept (PoC), using commercial equipment from overseas companies to secure data. The team is also working with affiliated companies to widely implement related technologies within Hyundai Glovis. In addition, the team plans to use CCTVs to analyze the movement, location, and behavioral relationships of recognized objects. These analyses will then be used to assess risks at the workplace and establish safety policies. Text recognition technology will be used by the team for inventory inspections and to improve recognition performance at logistics warehouses. In the case of drone inventory inspections, the team is working with overseas C/C to widely implement PoC. Improved QR recognition technologies are also to be used for parts recognition along the KD automation lines.

“The AI Tech team will strive not only to grow but also achieve meaningful results. To do this, one person trying hard alone is not enough. We ask for your continued support and interest.”

– Machine Vision Senior Manager Kim Seok-yeon

Parts supply chain optimization and automation equipment
Domestic & overseas CC

The AI Tech Team is developing models to optimize the supply chain of parts from domestic KD centers to overseas C/C and to improve the operational efficiency of automated equipment. Models are being developed to optimize KD center combination packaging, container loading, and automatic warehouse (ASRS) operations and to improve C/C container transportation scheduling and the handling and sorting of instruction models.

The key technologies required for the completion of these tasks are machine learning, optimization, and statistics. Machine learning in particular is critical for the implementation of AI algorithms and as such is used to research algorithms. During this process, the computer “learns” based on patterns and inferences, without being given explicit instructions by an outside user, and performs predictions based on its own data analysis.

Currently, the AI Tech Team is focused on developing and advancing optimization models for ASRS operation and container transportation scheduling. The ASRS optimization model decides where to store items and when and how to rearrange items. It also determines the robot’s paths of movement when retrieving items in order to process the maximum amount of cargo using a minimal amount of ASRS space. The container transportation scheduling model proposes the optimal C/C transportation schedule for containers stored in ports and container yards and containers undergoing maritime transportation. In the development stage, the team creates rule-based models using information gained through in-person interviews. Data is being collected to develop models that provide optimal transportation scheduling.

“Let’s give it our best this year!”

– Domestic & Overseas CC Senior Manager Yeon Han-byul

Developing unit tasks for the efficiency of KD centers
KD centers

The AI Tech Team is responsible for the optimization, prediction, natural language processing (NLP), recommendations, and other tasks for the smartification of the company’s KD centers. The overall goal is to maximize efficiency throughout the entire process at the KD centers, from the arrival of parts to their storage, picking, packaging, and release.

The team is developing algorithms for the KD centers that allows for the optimal packaging of various smaller boxes into larger boxes and the packaging of larger boxes into containers. Models are also being developed by the team to: suggest optimal packaging arrangements for new parts; analyze packaging methods (natural language) and packaging photos (images), etc. to classify and determine standard MH; develop algorithms to streamline ASRS operations; and create designs, and resulting policies, for the efficient operation of pallet lanes. This optimization is being achieved using the AI technologies such as reinforcement learning, prediction using multi-layer perceptron, and natural language processing using transformers.

Since there are many PoC-type tasks used to evaluate the technology level needed for the smartification of KD centers, it is often difficult to produce quantitative results. However, the AI Tech Team plans to create an overall “smart KD process” by optimizing each individual task.

“Although we clearly need AI, I don’t believe it is the answer for everything. There are still many things that need human intervention. Don’t think that AI will do everything, otherwise you’ll be disappointed. Instead, support the steady growth of AI within the company and cheer us on. We will work hard and do our best so that our company can become digitally smarter.”

– KD Center Senior Manager Jeong Hyun-hee

Developing smart pricing model for used cars
Auto-biz pricing

The AI Tech team is responsible for developing and operating the auto-biz pricing ML/AI model within the Trading Business Subdivision. They have already developed an ML/AI model for calculating secondhand car prices and are providing price information by linking internal systems, such as task-based systems and pricing systems. The model developed by the team also provides pricing information through external services such as SOCAR and Hyundai Capital and offers services to general customers through Naver My Car and AutoBell.

One of the team’s major achievements in 2022 was applying for two joint patents based on the outputs from the initial development phase of the pricing model. The team applied for the patents because their model signified the first time such a model was used for the calculation of used car prices. In order to utilize both detailed and more generalized factors for the prediction of vehicle prices, a hierarchical individual model was built and ensembled by applying multiple algorithms for each classifier. Finally, a 2-step model structure was developed to fit the previously calculated values.

Currently, the AI Tech Team is focusing on tracking accuracy and indicators to develop a consistent and stable pricing AI model, as operational environments continue to evolve. The team is working to improve model accuracy by conducting research, gaining insights from current AI models, applying new algorithms, and adding data features. Major tasks for this year include increasing the accuracy of the auto-biz pricing prediction model and creating dashboards for more convenient model operation. To enhance model accuracy, the team is considering adding external indicators for the used car market and making changes to the algorithm itself. To create an effective operational dashboard, actual workers out in the field need to be able to view many different components for each vehicle. It is also important to be able to check for different events and to check inventory. Currently, the team is sharing its progress in the form of a report twice a month.

“You are welcome to contact us any time if you have any questions about the processing of large amounts of data, other inquiries, or are simply wondering if something can be approached from an AI perspective. We will continue to work hard to make improvements in all areas.”

– Auto-Biz Pricing Senior Manager Hyun Won-jae

Q. Please tell us about yourselves.

Hi. I joined Hyundai Glovis in May 2022, and am responsible for planning and developing AI technology using images.

Hello. I joined the Industry Analysis Team at the Comprehensive Logistics Research Institute, now known as the Future Innovation Technology Center, in October 2020. I worked as part of my team’s Big Data WG and developed models for 3D bin packing and ship fuel price prediction. In July 2021, I was transferred to the AI Tech Team which was then known as the Big Data & AI Team and have been developing the AI and optimization technologies needed for Hyundai Glovis businesses.

I also joined the Industry Analysis Team’s Big Data WG in May 2019. Since joining the team, I have been working on tasks such as big data analysis and ML, and AI model development.

I guess I joined the company last. I started working here in August 2022, so it’s been eight months. I’m responsible for optimization, prediction, NLP, relevant recommendations, and other tasks related to the smart conversion of KD centers.

Q. You’re involved in tasks related to digital transition, such as automation and smartification. What is the most appealing thing to you about your work?

I think the thing that is the most appealing is that I always have new experiences even though I work on the same types of tasks. I particularly enjoy the feeling I get when I first encounter vast amounts of data and the problem that needs to be solved. Discovering patterns by looking at unfamiliar data and finding clues to a problem feels like exploring a whole new world. I feel a great sense of achievement when one of my hypotheses is proven right by the data and leads to the development of a model.

Machine vision tasks are based on images, so for every stage of my work, I can visually check the results, which, when the model is working properly, gives me a greater sense of achievement than other technologies. Also, the technology has a high degree of applicability and scalability, so it can be used in collaboration with many different business divisions instead of being limited to a specific business unit.

Q. Aren’t there lots of trials and errors and other difficulties in your tasks compared to other types of tasks?

First of all, we didn’t originally have a GPU development environment, so we started building an AI platform last October, and it was completed in March of this year. For now, only our team is using this platform because we still don’t have enough resources to go around, but we plan to gradually expand our resources so that the platform can be used by other departments as well. Since we’ve now secured the foundation needed for AI development, we’re working to achieve even better results.

I try to prioritize the overall direction of the department requesting the data analysis or AI model. Since the direction of the relevant department is the main reason for the task, whatever we develop must be aligned with the direction of the department in order for the thing we’ve developed to be used continuously.

In my case, I feel as if I hit a wall when there are differences between a model’s test environment and the operational environment. Identifying the causes for these differences is difficult because there are so many possible causes, but it is a crucial step.

There are many PoC-type tasks that involve inspecting the level of the technology or attempting to apply it to actual work tasks. So, there are many things that make it difficult to derive quantitative results. I hope that everyone in the workplace comes to the realization that AI technology needs to be performed in stages.

Q. What area do you pay the most attention to in terms of business direction?

I focus on quantifying and measuring the effects of operating the models developed by the AI Tech Team. To this end, I’m currently talking with the relevant department managers.

I’m also focusing on how to set quantitative and qualitative indicators to measure performance.

I think it’s very important to determine the scope of applicability for the latest technologies by identifying the needs of the business divisions. It is natural for people to be hesitant about actually applying new equipment or technologies. When I try to determine the best way forward, I take into consideration the needs of the business divisions and the current level of the technology and verify the reliability of the technology in stages.

Q. What are the criteria you use to identify AI tasks that meet the needs of each business division?

The AI Tech Team holds project briefings for each business division in the 3rd or 4th quarter of every year, and at these briefings, we discuss tasks we need to work on in the following year. We also discuss projects for each division in terms of degree of business contribution, scope of technology utilization, and the feasibility of execution.

We check the current level of the AI technology to see if the needs out in the field align with the AI’s capabilities and then decide whether the technology can be applied.

The urgency of each project is also a key criterion. If a task is essential to the business division and requires immediate action, it is made a priority for review. These potential projects are reviewed in terms of data, methodology, and feasibility in the actual work environment.

Q. What was your proudest achievements that you accomplished through digital transition?

I built an AI platform to create a development-friendly environment for our team members. We expect that many more programs will be developed using this platform. This year, I plan to work on visionary projects for greater results.

We’ve been able to achieve great results in terms of our prediction of ship fuel prices, KD combination packaging, container loading optimization, container scheduling, and handling and sorting. I feel rewarded when people use the models we developed in their work.

I feel most rewarded when a model we developed is interfaced with an existing system and actually used for business operations. Although there are many cases in which we perform short-term prediction tasks, I feel the most fulfilled when a model we’ve developed is used expanded and used long-term. It’s also rewarding when we implement different ideas for variable processing and algorithm application during the model development phase and the predictive power of the model improves.

We applied in-depth technology, such as reinforcement learning, to KD combination packaging optimization and container loading automation and were able to accumulate a great deal of know-how. Since I have only worked at Hyundai Glovis for only a short amount of time, I’d like to have an even greater sense of pride through solid achievements this year.

Team Leader Jeon Ji-won

Q. What are your future plans and goals?

The AI Tech Team is actually not even two years old yet. It’s only been a short amount of time, but we now have 12 team members and are working on over 20 projects. In terms of the development environment, since we have built our AI platform, I feel like our team is just at the starting line. Currently, we are working on securing data and PoC for unit technology verification as a part of our vision project. It takes a great deal of time to apply a technology so that it actually works. For now, our technology is still at the stage of supporting human judgment, but once enough data has been collected and the AI can use this data to learn, it will be able to replace humans. In fact, there are over 100 million confirmed cases of visual inspections being performed using overseas commercial technology, so in comparison, the AI Tech Team’s vision project is still in its infancy. We plan to apply our technology internally by benchmarking commercial technologies and creating a mid- to long-term technology roadmap that will allow us to develop and apply our own technologies.

I plan to develop necessary models to establish a connected SCM that can achieve maximum operational efficiency using a minimal amount of personnel and space through the optimization of the KD supply chain, connecting domestic KD centers, maritime transportation, external C/C, and automobile factories. I also plan to study recent trends in deep learning research, which is something I’ve been putting off with the excuse of being too busy, and further enhance my professional expertise.

AI technology is rapidly developing, so if I don’t continually study it, it gets difficult to apply the technologies that are commercially available in the market. I plan to learn about new and different technologies to make our KD centers smarter as well as to extensively apply different technologies to achieve meaningful results.

I’m planning to create a new deep model for our auto-biz pricing and related pricing AI models, as well as to continue researching system API I/F and feature engineering. I hope to contribute to expanding our used car business model through advanced pricing calculations.

Q. What is your long-term plan and vision for work automation and optimization?

The ultimate goal of the vision project is, using the programs created by Hyundai Glovis, to automate everything that is seen and worked on in order to increase our overall productivity. It looks like it will be a very long journey.

My goal is to provide solutions and develop and commercialize necessary technologies to automate as many of the domestic KD centers and outside C/C operations as possible and to maximize the operating efficiency of our automation equipment.

Although we have implemented partial automation everywhere, from our warehousing to shipping in KD centers, I believe we still have a long way to go before we can achieve full automation. However, our goal is clear. We plan to create a smart KD in which the optimization of individual tasks does not get in the way of optimizing the overall workflow.

Even though it’s not a long-term goal, my main goal is to further develop our pricing AI model. By upgrading this model, we can cover a broader range of vehicle models and create a standard model for vehicle purchases by business operators. Our AI model employs a concept that allows for the automation of more extensive areas and tasks related to the used car business. I also hope to develop and operate advanced algorithms that can calculate prices that are more competitive than those calculated by other companies, so that our pricing services will be the best and most sought after in the used car market.

By the Editorial Department