Neural Networks and the Future of Mankind
What used to be a scene out of a sci-fi movie is now within our technological reach—cutting-edge technologies in Artificial Intelligence (AI), such as Deep Learning and Natural Language Processing (NLP), have evolved to the point that they influence our everyday life. Which naturally leads us to ask: what are their ramifications on the future of mobility?
Pursuing the vision of becoming a smart mobility solution provider, Hyundai Motor Group is developing AI-based technologies that provide innovative mobility services for our customers and optimize the production process. From an AI assistant who supports its driver to technology that renders optimized route calculation, and even to smart factories that take on the difficult tasks on behalf of people, in the next few paragraphs, we introduce some of our efforts in the new age of mobility.
FUTUREUnprecedented Car Life, Smart Factory
– the Key is in Data Intelligence
For the industries of the future, AI will undoubtedly be the main pillar; its wide applicability to many fields—including voice recognition, manufacture/production, and autonomous driving, to name just a few aspects relating to the automotive industry—represents a huge potential for human progress. Seeing this, Hyundai Motor Group established in 2019 the AIRS Company (formerly the AIR Lab), a dedicated organization exclusively for AI technologies and applications. The company is currently working in three domains: automotive voice agent, mobility service platform, and smart factory.
The easiest analogy to the in-car voice agent would be Jarvis, a versatile AI agent from Marvel’s hit series Iron Man. Of course, our technology is not quite as advanced yet, but the basic principle—of working ‘smart’ to meet even the unspoken needs of the user—is the same. As the self-driving car runs and parks by itself, the driver may enjoy the assistance of the AI Agent, who, upon simple voice command, guides the driver through the infotainment system, media contents, and even basic life services. Sometimes, the AI Agent may become active even when there is no explicit command; almost like a skilled butler, it will actively perceive the user’s need and request permission to fulfill that need.
Meanwhile, the basic purpose of a car—transportation—will become more efficient with artificial intelligence in the mix. For example, as of now, the navigation’s route calculation does not fully accommodate multiple destinations. But with artificial intelligence, several passengers with different destinations will be able to use a single optimized, seamless route. This technology was certified in 2020 by the on-demand mobility services ‘Shucle’ and ‘I-MOD.’
Finally, the Group is working hard to improve the foundational infrastructure through which its vehicles are assembled—the factories of yesteryear are being renovated into “Smart Factories.” A smart factory is powered by intelligent robots, capable of making human-like judgments, that assist or replace humans in tiresome, repetitive tasks. A smart factory should be better geared to handle small quantity batch production, production of multiple different model lines by small amounts in the same production facility. The consumers will in turn benefit by having the ability to find a customized product, one that best fits their preferences, at a low price but with high quality.
KEY TECH1. Automotive Lifestyle Technology
Until now, speech recognition in cars could only respond to very simple commands. For example, the driver may say, “Turn on the A/C.” The system would process this command and turn the A/C on, end of story. But the AI Agent being developed by Hyundai’s AIRS Company functions at a higher level, taking into account the current status of the vehicle and the driver and providing services that understand the motivation behind the spoken command. For example, “Turn on the A/C” should be followed up with another query, “Should I close the windows then?” The “open the windows” command during a heavily polluted day should prompt another clarification: “The PM level outside is unsafe. Should I open the windows anyway?” In the same context, if the driver is detected as potentially dozing while driving, the vehicle might automatically pull the windows down to bring in fresh air or even ask the driver if he or she would like to take a rest.
These capabilities naturally require an understanding of the driver’s intent underneath his command; Hyundai Motor Group is investing in the R&D of the NLU (Natural Language Understanding) system and other technologies that facilitate AI-driven conversations.
No need to go on to the internet (or the hefty manual book) anymore for answers to questions regarding your vehicle—just ask your car instead. The ‘Hyundai QnA,’ an AI-based FAQ system, can respond to such questions as “What is ADAS?” and “Tell me some tips for car maintenance during summer.” It will give answers tailored to the vehicle specifications in a manner understandable by people at any level of expertise. Thanks to its exhaustive database of knowledge, the system can also serve as a comparison tool for people looking to buy new cars that best match their criteria.
Hyundai Motor Group’s present goal is to improve the system to respond not just to car-related questions but also to those on traffic law or insurance claims. Once the system’s knowledge database is sufficiently developed, the next step would be to expand its language capabilities for global distribution.
2. Mobility Technology
Dynamic Routing Algorithm
This AI-driven algorithm works for on-demand mobility services where vehicles owned by the service need to be routed and allocated to the users efficiently. The algorithm analyzes the optimal route by constantly referring to real-time traffic and road data, which allows it to simultaneously serve multiple passengers with different destinations. It has been found to minimize the service vehicles’ standby and movement time by significant amounts. The algorithm was adopted by the on-demand mobility service Shucle, which tested successfully in the market in early 2020.
Passenger Detection Module
In certain public transport services like long-distance bus chains, the driver has the responsibility of manually checking whether the passengers are sitting according to the reservation record. This arrangement is inconvenient, not to mention imprecise, as evidenced by constant security or free-rider problems that these services experience. The Passenger Detection Module uses the built-in-vehicle RGB camera and an AI-based seating detection software to automatically display the current seating status, allowing the driver to focus just on the driving. Conveniently, the software can be modeled to detect free-riders or passengers without reservations as well.
3. Smart Factory Technology
Deep-Learning Scans of Paint Inspection Sheets
This technology takes advantage of a large existing database of paint inspection sheets, which are manual records of painting quality inspections by human inspectors. The information in the sheets (inspection time, vehicle model, problem type, problem locations) is scanned to build a Big Data database that helps engineers better address the problems that appear in a pattern or are specific to a certain model. Hyundai Motor Group is planning to expand this mechanism to other quality control or production processes like underbody inspections and finished-vehicle visual inspections.
Automatic Wheel Alignment Calibration Technology
Wheel alignment, an angle adjustment of a vehicle’s suspension, is periodically required to keep the vehicle running precisely straight. Improperly aligned wheels can lead the vehicle to veer off-center even when the driver is steering firmly straight. The automatic wheel alignment calibration technology uses Deep Learning to assist in the alignment process. The AI learns from the past wheel alignment data, predicting the ideal adjustment values for each vehicle model; these predicted values are instructed to the machine, which completes the alignment job; the job results are then returned to the AI, which learns from its errors and seeks to reduce the error value, in a reiterating cycle of trial-and-error that helps the system “learn” to be more precise. Once it has “learned” enough, the AI can accurately predict optimal adjustment values even when a completely new wheel alignment angle is given.
Wheel Design Generator
AI helps not only the production process but also vehicle design. The Wheel Design Generator, a recent R&D project of Hyundai Motor Group, uses an AI algorithm to create wheel patterns out of everyday images that the designers want to use as inspiration. The AI-generated wheel patterns can essentially serve as the ‘midpoint’ from which the designers can do the fine-tuning, an arrangement that drastically reduces the time and effort required to develop new designs.
4. Work Support
Hyundai Translator (App/Web)
Hyundai Translator is an intra-company portable translation program for Hyundai Motor Group workers, whose global communication and travel duties require an efficient means to translate across languages. It is capable of translating speech in one-on-one or multichannel conversations as well as translating images. The pre-existing translation software required users to copy-and-paste the text on to the service page in order to receive the translations; however, the Hyundai Translator, integrated within the intra-company messenger and mail service, eliminates this process and makes it all-in-all easier for Hyundai workers. Another advantage that sets it apart from the competition is that an ongoing research is being done to tailor the API (Application Programming Interface: a software intermediary that allows the application and the OS to interact with each other) to better address security issues.
5. Sports Support
‘AI Coach’ for the National Archery Team
Based on Deep Learning Vision technology, the AI Coach is an automatic video editing application that produces individualized training/competition videos that help identify the archer’s weaknesses. The program targets only the highlights from one long take and builds datasets for two ends: one for viewing the postures and another for viewing the targets. The program then syncs the posture and the target clips into one simultaneous stream, which helps the coaches and the players to better identify bad habits and/or weaknesses. Because the program can identify even the slightest posture differences invisible to the naked eye, it helps the Korean National Archery Team in maintaining its world-class stature.