Deep Learning Helicopter Dynamics Models（深度学习直升机动力学模型）.pdfVIP

Deep Learning Helicopter Dynamics Models Ali Punjani and Pieter Abbeel Abstract— We consider the problem of system identification of helicopter dynamics. Helicopters are complex systems, cou- pling rigid body dynamics with aerodynamics, engine dynam- ics, vibration, and other phenomena. Resultantly, they pose a challenging system identification problem, especially when considering non-stationary flight regimes. We pose the dynamics modeling problem as direct high- dimensional regression, and take inspiration from recent results in Deep Learning to represent the helicopter dynamics with a Fig. 1. The dynamical modeling method presented in this work is used to Rectified Linear Unit (ReLU) Network Model, a hierarchical model a helicopter’s dynamics as it is flown through aerobatics. A unified neural network model. We provide a simple method for ini- model is learned for a rich array of maneuvers, including flips, loops, tializing the parameters of the model, and optimization details stop and go, vertical sweeps, circles, dodging, orientation sweeps, gentle freestyle, tictocs, chaos, and aggressive freestyle. for training. We describe three baseline models and show that they are significantly outperformed by the ReLU Network aerobatics, however, have only been successfully flown under Model in experiments on real data, indicating the power of the autonomous control using Apprenticeship Learning [11]. In model to capture useful structure in system dynamics across the Apprenticeship Learning approach for autonomous hel