전형석 모라이 이사는 모라이와 앤시스가 공동 개최한 ‘KADF 2023(Korea Autonomous Developer Forum)’에서 ‘자율주행 시뮬레이터를 활용한 가상 센서 데이터 응용 사례’에 대해 발표하며 자율주행 기술 평가 지표에 대해 논점을 던졌다.

▲Director Jeon Hyeong-seok of Morai is giving a presentation.
“Autonomous driving on highways and alleyways should be evaluated by different indicators”
Style-transferred deep learning provides customized virtual data Jeon Hyeong-seok, CEO of Morai, said that there is a need to consider evaluation criteria that can answer the question of which vehicles are good at autonomous driving.
At the 'KADF 2023 (Korea Autonomous Developer Forum)' co-hosted by Morai and Ansys, Jeonhyeong Seok, Director of Morai, presented 'Application Cases of Virtual Sensor Data Using Autonomous Driving Simulators' and raised a point about autonomous driving technology evaluation indicators.
Currently, the system evaluation for autonomous driving uses driving distance and frequency of disengagement as indicators.
Director Jeon Hyeong-seok questioned whether a company should be recognized as an excellent autonomous driving company just because it ranks high in the two indicators above.
The opinion is that if testing is only done on highways, autonomous driving development should be evaluated using completely different indicators.
Former CEO Jeon stated his thoughts, saying, “Thinking about any evaluation indicators for autonomous driving systems is a form of research and something that needs to be thought about.”
He is an evaluation metric for virtual data sets.Likewise, autonomous driving requires data appropriate to the environment because it drives in various environments.
One way to overcome these challenges is to use style-transferred deep learning techniques to provide virtual data tailored to the user's desired environment.
Former CEO said, “We internally confirmed that it is much more useful to use virtual data from simulations that has been filtered or style-transferred to have similar characteristics to actual data rather than directly using sensor images and data from Morai and Ansys solutions.”
He continued by saying that this is a transitional period to an environment where all vehicles are autonomous, and at the same time, it is the most difficult time, and that the range of data sets required is bound to be very broad because the characteristics of drivers surrounding autonomous vehicles are very diverse.
Former CEO said, “I am in a position to lead the development of autonomous driving technology,” and added, “I expect that if we all work together toward a unified goal, we will be able to develop better autonomous driving technologies.”
Meanwhile, former CEO Jeon expressed his doubts about whether the data sets currently being distributed or built by companies and institutions are truly general.
The view is that there has been no standardization in the types, numbers, or locations of sensors attached during vehicle development, and that data acquired from different configurations by each company is useless.
In addition, although data for ordinary situations can be acquired very easily, there is a disadvantage in that there is little data for edge cases and difficult events that autonomous driving ultimately has to solve.
If a single problem occurs on the road, trust in autonomous driving technology could be undermined.There is no outside.
Morai said that in order to resolve this situation, the use of a simulator is necessary, so Morai and Ansys collaborated to create a dataset that leverages each company's strengths, and built a large-scale dataset with approximately 38 million samples.