Machine Learning for Helicopter Dynamics Models

摘要: Machine Learning for Helicopter Dynamics Models by Ali Punjani Master of Science in Computer University California, Berkeley Professor Pieter Abbeel, Chair We consider the problem system identification helicopter dynamics. Helicopters are complex systems, coupling rigid body dynamics with aerodynamics, engine dynamics, vibration, and other phenomena. Resultantly, they pose a challenging problem, especially when considering non-stationary flight regimes. modeling as direct high-dimensional regression, take inspiration from recent results Deep to represent Rectified Linear Unit (ReLU) Network Model, hierarchical neural network model. provide simple method initializing parameters model, optimization details training. describe three baseline models show that significantly outperformed ReLU Model experiments on real data, indicating power model capture useful structure across rich array aerobatic maneuvers. Specifically, improves 58% overall RMS acceleration prediction over state-of-the-art methods. Predicting along helicopter’s up-down axis is empirically found be most difficult, 60% prior state-ofthe-art. discuss explanations these performance gains, also investigate impact hyperparameters novel