The rapid advancement of intelligent vehicle technology brings new challenges in control and state estimation, particularly in extreme scenarios where traditional methods may be insufficient. This dissertation addresses these challenges by developing novel algorithms for autonomous drift system and attitude estimation, with a focus on enhancing vehicle safety and performance under challenging conditions. The proposed solutions are extensively validated through simulation or real-world experiments, demonstrating their effectiveness across diverse operating environments.

First, we propose a Deep Reinforcement Learning based Drift Obstacle Avoidance (DOA) system for intelligent vehicles operating in emergency situations. The system leverages deep learning techniques to execute controlled...[ Read more ]

The rapid advancement of intelligent vehicle technology brings new challenges in control and state estimation, particularly in extreme scenarios where traditional methods may be insufficient. This dissertation addresses these challenges by developing novel algorithms for autonomous drift system and attitude estimation, with a focus on enhancing vehicle safety and performance under challenging conditions. The proposed solutions are extensively validated through simulation or real-world experiments, demonstrating their effectiveness across diverse operating environments.

First, we propose a Deep Reinforcement Learning based Drift Obstacle Avoidance (DOA) system for intelligent vehicles operating in emergency situations. The system leverages deep learning techniques to execute controlled drift maneuvers across diverse conditions, demonstrating remarkable capabilities in extreme scenarios where conventional obstacle avoidance methods may fail. Our extensive evaluations show that the DOA system achieves significantly improved obstacle avoidance performance while maintaining vehicle stability. Second, we introduce two complementary approaches to attitude estimation: a novel biquaternion-based framework that enhances estimation accuracy and convergence performance, and a simplified closed-form solution for GNSS/accelerometer/magnetometer integration. Both methods achieve superior performance while maintaining computational efficiency on embedded platforms, making them particularly suitable for resource-constrained autonomous systems.

The experimental results demonstrate that our proposed methods effectively enhance vehicle safety and performance in challenging scenarios. The integration of these components provides a comprehensive solution for intelligent vehicle control and state estimation in extreme conditions. Our research contributes significantly to the advancement of intelligent vehicle technology, offering new possibilities for improving vehicle safety systems in scenarios where traditional approaches may be inadequate. These methods have been successfully tested, demonstrating their practical applicability and robustness in autonomous driving and driving assistance applications.