
AI Perception Model Development
AI Perception Model Development
We develop AI perception models from the ground up, starting with architecture design, optimized for automotive SoCs and compute-constrained edge devices. Our work covers image recognition tasks for AD/ADAS (autonomous driving and advanced driver-assistance systems), including object detection, lane detection, and traffic sign and traffic light recognition. We also compress the latest architectures, such as Vision-Language-Action (VLA) models, for efficient deployment. We take on the entire process, from R&D to mass-production deployment.

Key Techniques
- Knowledge distillation from large models into lightweight models for on-device deployment
- Multi-task model architectures designed around target inference cost and SoC characteristics
- Pseudo-labeling with large models and image retrieval systems for more efficient development
- Training data enrichment through world-model-based data generation and scene augmentation
- Integration with adjacent technologies, such as camera pose estimation using IMU and vehicle speed sensors
FEATURES
Perception Models Built In-House, from Scratch
We handle every stage of perception model development in-house, from design and training to accuracy improvement. Our models are purpose-built for automotive use cases such as object detection, lane detection, traffic sign recognition, and traffic light recognition, delivering both the accuracy and the processing speed your product requires.

Ready for the Latest Model Architectures
Beyond core tasks such as object detection, lane detection, and traffic sign and traffic light recognition, we take on advanced tasks, including open-vocabulary object detection and scene understanding, as well as the latest architectures such as Vision-Language-Action (VLA) models. We are also working on compressing reasoning-capable, VLA-based autonomous driving models, turning advances in the fast-moving AI field into solutions that can be deployed in automotive products.

From Research to Mass Production
We have the expertise to adapt the latest AI technologies to automotive-specific constraints: cost, power consumption, and real-time performance. Our team includes engineers who have worked on mass-production programs at OEMs and Tier 1 suppliers, and we support every step of turning research results into production-ready products.
