Hybrid Neural Network and Conventional Controller Approach for Robust Control of Highly Unstable Systems: Application to Tilt-Rotor Control
Researchers have developed a novel control system for tilt-rotor drones, which are known for their advanced maneuverability but also their inherent instability. Initial attempts using direct neural network control with MLPs, LSTMs, and transformers proved unsuccessful in stabilizing the system. The team's main contribution is a hybrid approach that combines a neural network with a sliding mode controller, where lightweight networks learn specific system dynamics from flight logs, significantly improving robustness and reducing computational load. AI
IMPACT This hybrid control system could enable more stable and robust operation of advanced drones in complex environments.