The Correlation between Vehicle Vertical Dynamics and Deep Learning-Based Visual Target State Estimation: A...

Area of Science:

• Automotive Engineering
• Artificial Intelligence
• Robotics

Background:

• Automated vehicles promise enhanced transport convenience, mobility, and reduced congestion.
• Safety evaluation and validation remain the primary obstacles to widespread automated vehicle deployment.
• Artificial Intelligence (AI)-based vision systems are critical for automated vehicle perception and safety.

Purpose of the Study:

• To analyze the influence of vertical vehicle dynamics, specifically road anomalies and suspension, on object detection and tracking performance.
• To identify limitations in AI-based vision systems under varying dynamic conditions.
• To provide insights for improving the safety validation of automated vehicles.

Main Methods:

• Conducted an extensive road field study to collect real-world data.
• Validated a computational tool for assessing AI vision performance using simulations.
• Performed a parametric study to evaluate AI vision under different vertical dynamics scenarios.

Main Results:

• Vertical vehicle dynamics, influenced by road anomalies and suspension, demonstrably affect AI vision performance in detecting and tracking objects.
• Identified specific conditions where AI-based vision underperforms, potentially compromising automated vehicle safety.
• The study highlights the critical need to consider vertical dynamics in AV perception system design and testing.

Conclusions:

• Vertical vehicle dynamics are a crucial, previously overlooked factor in automated vehicle safety validation.
• Current AI vision systems may exhibit performance degradation under certain dynamic road conditions.
• Further research and development are needed to enhance the robustness of AI vision for safe automated vehicle operation across diverse environments.