Learning quadrotor dynamics using neural network for flight control

Abstract

Abstract-Traditional learning approaches proposed for controlling quadrotors or helicopters have focused on improving performance for specific trajectories by iteratively improving upon a nominal controller, for example learning from demonstrations, iterative learning, and reinforcement learning. In these schemes, however, it is not clear how the information gathered from the training trajectories can be used to synthesize controllers for more general trajectories. Recently, the efficacy of deep learning in inferring helicopter dynamics has been shown. Motivated by the generalization capability of deep learning, this paper investigates whether a neural network based dynamics model can be employed to synthesize control for trajectories different than those used for training. To test this, we learn a quadrotor dynamics model using only translational and only rotational training trajectories, each of which can be controlled independently, and then use it to simultaneously control the yaw and position of a quadrotor, which is non-trivial because of nonlinear couplings between the two motions. We validate our approach in experiments on a quadrotor testbed.

Neural Lander: Stable Drone Landing Control Using Learned Dynamics

“…It can be shown that this leads to finite-gain L p stability and input-to-state stability (ISS) [29]. Furthermore, the hierarchical combination between s andp in (7) results in lim t→∞ p(t) = lim t→∞ s(t) /λ min (Λ), yielding (14).…”

“…where the reference angular rate ω r is designed similar to (7), so that when ω → ω r , exponential trajectory tracking of a desired attitude R d (t) is guaranteed within some bounded error in the presence of bounded disturbance torques.…”

Low-Level Control of a Quadrotor With Deep Model-Based Reinforcement Learning

“…Researchers of system identification for quadrotors also apply machine learning techniques. Bansal et al used NN models of the Crazyflie's dynamics to plan trajectories [5]. Our implementation differs by directly predicting change in attitude with on-board IMU measurements and motor voltages, rather than predicting with global, motion-capture state measurements and thrust targets for the internal PIDs.…”

“…This paper is the first demonstration of controlling a quadrotor with direct motor assignments sent from a MBRL derived controller learning only via experience. Our work differs from recent progress in MBRL with quadrotors by exclusively using experimental data and focusing on low level control, while related applications of learning with quadrotors employ low-level control generated in simulation [4] or use a dynamics model learned via experience to command on-board controllers [5]. Our MBRL solution, outlined in Figure 1, employs neural networks (NN) to learn a forwards dynamics model coupled with a 'random shooter' MPC, which can be efficiently parallelized on a graphic processing unit (GPU) to execute low-level, real-time control.…”

Closed-loop Model Selection for Kernel-based Models using Bayesian Optimization

“…This problem has been well studied in literature for both linear systems, e.g. [5], as well as for nonlinear systems using the function approximators such as GP [6]- [8] and neural networks (NN) [9], [10]. However, a model obtained using this open-loop procedure can result in a reduced controller performance on the actual system [4].…”