Researchers have developed a novel method to learn the regions of attraction for transverse feedback linearization (TFL) controllers, which are crucial for ensuring stable operation in systems like unicycles and quadrotors. The proposed approach decomposes the value function into an analytically known quadratic cost-to-go and a learned safety value function, which is trained using fitted value iteration without requiring infinite-horizon trajectory labels. This method significantly outperforms existing techniques, such as physics-informed neural networks, in accurately estimating the region of attraction across various simulated scenarios. AI
IMPACT Enhances the safety and reliability of autonomous systems by improving controller stability analysis.
RANK_REASON Academic paper detailing a new method for learning control system properties. [lever_c_demoted from research: ic=1 ai=1.0]
- Fitted Value Iteration
- Hugging Face
- interval branch and bound
- Physics-Informed Neural Network
- quadcopter
- Transverse Feedback Linearization
- unicycle
- Zubov
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →