Four Pillars ofPerception.

Radar-Centric Sensor Fusion

We develop multimodal 3D fusion technology that understands the surrounding environment three-dimensionally by combining information from heterogeneous sensors, centering on 4D imaging radar along with cameras and LiDAR. We unify the range and velocity from 4D imaging radar, semantic data from cameras, and precise 3D spatial structures from LiDAR in a shared coordinate space. This generates highly precise 3D environmental data that cannot be achieved with single sensors. Our expandable fusion perception model supports stable spatial awareness regardless of changes in weather, illumination, or sensor states.

Virtual Radar-Based Synthetic Data Generation & AI Training Framework

We develop an integrated training platform that generates synthetic datasets within virtual sensor environments reflecting 4D imaging radar characteristics, utilizing them for AI model pretraining and development. We efficiently acquire training datasets by simulating diverse objects and operational conditions in virtual environments that are costly or difficult to collect in the real world. Models pretrained on synthetic data can be fine-tuned and optimized for real-world sensor data. Furthermore, we streamline actual data labeling and verification processes using auto-labeling based on pretrained models, supporting the end-to-end AI development pipeline from data building to model training and site optimization.