Advanced Data Augmentation for ADAS

The development of ADAS AI is often bottlenecked by "Edge Cases"—rare weather conditions, unique traffic signs, or unpredictable behaviours that are too costly or dangerous to capture in the real world. While fully synthetic data generated by game engines was a potential solution, the AI models were often confused by the "perfect" nature of these simulations. Additionally, every hardware or software iteration required fresh data that met strict environmental and geographical specifications.

To solve this, the team developed a sophisticated augmentation software that combines multiple AI technologies.

• Hybrid Enrichment: The goal was to enrich reality so precisely that the AI couldn't distinguish between real and synthetic data.

• Sensor Degradation: The tool realistically simulates environmental impacts on sensors, such as salt spray, mud, or heavy rain, without requiring physical field tests.

• Precision Augmentation: It allows for the placement of rare or geographically specific traffic signs into real footage across various lighting and weather conditions.

The project achieved a breakthrough in both performance and economics:

• Efficiency: AI models reached required confidence levels with significantly less data compared to traditional methods.

• Cost Savings: The tool created an annual saving potential of €2 million to €5 million.

• High ROI: The ongoing development cost of the tool is approximately 1/10th of the annual savings it generates.

• Strategic Alignment: Due to the tool's effectiveness, the company decided to focus on internal deployment to maximize its own competitive advantage in the autonomous driving sector.