Three Pillars Supporting the Goal of Practical Application
- Algorithms loaded into in-vehicle devices aim for the highest level of performance but must also meet the restriction that they can be applied practically as a product to meet the needs of the customer. In algorithm development, we continue to incorporate cutting-edge AI technology rapidly, and to provide sufficient learning data and computing environments to continue to achieve the acceleration and memory efficiency required for in-vehicle devices.
- In the development of semiconductors, we work with algorithm developers to produce high-level performance within the restrictions of in-vehicle products to develop advanced technology from the standpoint of miniaturization and acceleration. We have also achieved early practical application of AI systems in vehicles by obtaining the cooperation of semiconductor companies to rapidly advance in-vehicle projects.
- Quality Foundation
- It is essential that quality be built into every step in the development workflow. With the introduction of machine learning technologies, new technologies for quality assurance are necessary in various situations during development. As part of our quality foundation, we continue to develop core technologies to ensure the quality of learning data, the quality of learned models, and the quality of the systems using the models.
- Image Sensing
- trACE High-speed, Memory Efficient Landmark Point Estimation Technology
Image sensing is an essential technology for advance driving assistance. However, high-speed processing of multiple images obtained from cameras is required. As a result, we are working on basic logic research to reduce the amount of computation for image detection. The trACE high-speed, memory efficient landmark point estimation technology produced as a result of this research enables high-speed detection of every part of the subject (the tires, windshield, and hood for the car; the nose, eyes, mouth, etc. for a person's face). By running advanced image recognition technology on the in-vehicle computer, it is expected that the potential for advanced driver assistance will increase even further.
- Autonomous Driving Algorithms
- "What-If Driving" Taking into Consideration Various Risks
To support appropriate driving under complex conditions, such as city streets, and to make safe, smooth autonomous driving a reality, it is necessary to predict various risks and to determine optimal actions quickly. We are building artificial intelligence technology and machine learning algorithms to predict how driving situations can change and also developing real-time motion plan algorithms that can apply these predictive results and determine optimal actions quickly. Based on data accumulated in real vehicles, we have proceeded with technological validation on various platforms including simulated miniature cars.