Hi, We are LiOps
LiOps provides essential data curation for autonomous driving using the latest 3D deep learning technology.
We dream of a world where we achieve Level 5 autonomous driving through our services.
LiOps Story
"Why is it still difficult to commercialize autonomous driving beyond Level 3?"
To achieve highly autonomous driving, it is necessary to secure a vast amount of driving data and conduct direct experiments with autonomous driving logic developed based on this data. However, due to various regulations, it is almost impossible to conduct autonomous driving experiments in real-world situations where accidents are likely to occur.
LiOps is a deep tech startup that started with this question.
We have directly met with the ADAS development team of a well-known domestic OEM, an autonomous delivery robot startup, a university autonomous driving research lab, and a representative from the largest domestic robot arm company. Through this process, we have learned the following lessons:
✅ The most important aspect of autonomous driving is perceiving 3D spatial information from 2D images.
✅ A new data curation service is truly needed. Although there is a lot of refined data available on the internet, suitable data for each company's desired vehicle or robot ultimately needs to be acquired and processed directly, which incurs astronomical costs.
✅ It is difficult to collect corner cases. We need data on various environments and accident situations that are difficult to pre-train our autonomous driving AI on, such as fog and severe weather conditions. However, collecting this data in a wide range of real-world scenarios is almost impossible.
In the end, we decided to create a virtual data simulation based on reality.
3D vision-based autonomous driving is experiencing explosive growth
At the 2022 CVPR, Tesla discovered and began using occupancy prediction. This technology has been selected as one of the four disciplines in the autonomous driving competition at the 2023 CVPR. It has become a competitive technology where big tech companies and university research labs from various countries compete based on its performance.
Occupancy prediction, in simple terms, refers to the process of generating a result similar to the image on the right from the image on the left, enabling vehicles to avoid various objects in a 3D space. The closer the predicted results seen in the middle are to the ground truth, the higher the performance of autonomous driving.
LiOps is developing a service that can provide an unlimited amount of data required for AI training based on 3D vision.
“Why now, and Why LiOps?”
The reason Tesla was able to significantly improve performance using this technology is not only because of its self-driving technology, but also because it can receive a huge amount of data every day from millions of vehicles around the world to develop the technology. However, in the end, these data are only changes in external environmental variables such as time zone, traffic conditions, and weather in a given road environment. So, if we could reconstruct the entire region in 3D, starting from Korea's self-driving pilot district, and generate reality-based data by manipulating environmental variables with AI, wouldn't we be able to equally utilize Tesla's strength of big data?
This idea is something that only LiOps can think of and develop to date.
Experience working in the Hyundai Mobis autonomous driving team,
which was the first in Korea to introduce deep learning to autonomous driving,
AI developer with the global No. 1 record in LiDAR-based 3D segmentation
The first company in Korea to provide 3D deep learning-based reality reconstruction technology called NeRF as SaaS. Developers who worked as MLOps worked together and succeeded in creating a virtual reality that is identical to reality.
