Application of an approximate model predictive control scheme on an unmanned aerial vehicle

Abstract

An approximate model predictive control approach is applied on an unmanned aerial vehicle with limited computational resources. A novel method using a continuous time parametrization of the state and input trajectory is used to derive a compact description of the optimal control problem. Different first order methods for the online optimization are discussed in terms of memory requirements and execution time. The generalized fast dual gradient method is implemented on the aerial vehicle. The approximate model predictive control algorithm runs on an embedded platform with a STM32 Cortex M4 processor. Simulation studies show that the model predictive controller outperforms a linear quadratic regulator in aggressive maneuvers. The model predictive control approach is evaluated in practice and shown to yield satisfactory flight behavior.

Experience-driven Predictive Control with Robust Constraint Satisfaction under Time-Varying State Uncertainty

“…Therefore, given a set of active constraints, the corresponding controller and KKT matrices can be reconstructed online using (13), (15), and the current A, B,c, δ…”

“…That is, we require the ability to compute predictive control commands at sufficiently high rates to ensure stability of these agile systems. Fast MPC solution strategies can be divided into four categories: leveraging fast online optimization techniques [26], optimizing approximate formulations [15], explicit enumeration of equivalent controllers [2], and semi-explicit approaches [7,8,28]. In this work, we consider this last class of techniques due to the reduced reliance on online optimization in a critical control loop and their scalability to available computational resources [28].…”

“…As this algorithm runs, M will be populated with the appropriate controllers for the current situation, thereby reducing the dependence on the intermediate controller. Due to the parameterized form of the controller gains (13) and KKT matrices (15), the elements of M also automatically adapt to changes in the dynamics model and robustness bounds, thus maintaining robust constraint satisfaction via controller switching. Finally, we note that switching controllers within the database preserves stability as it is analogous to explicit MPC techniques [13], while transitions to and from the intermediate controller will preserve stability if they are sufficiently infrequent [6,14].…”

A direct visual servoing‐based framework for the 2016 IROS Autonomous Drone Racing Challenge

“…1 These systems enabled many impressive research results that were not previously possible owing to limitations such as the weight, accuracy, or the dynamic range of the onboard sensor. [2][3][4][5][6][7][8][9] However, such delicate and expensive systems cannot be assumed always available in real world, to deploy UAVs into various real-world applications, one needs to employ a self-contained navigation method. Among a few solutions, the vision-based navigation is quite promising: indeed, a variety of vision-based approaches ranging from target recognition to visual servoing [10][11][12][13] have been successfully applied to UAVs.…”

Novel Levenberg–Marquardt based learning algorithm for unmanned aerial vehicles

“…Proportional-integral-derivative (PID) (Eresen et al, 2012), dynamic feedback linearization (DFL) (Hua et al, 2013), linear-quadratic regulator (LQR) (Dong et al, 2015), and model predictive control (MPC) (Hofer et al, 2016) are the examples of the most widely used model-based controllers. The aforementioned controllers deliver a good balance between implementation cost, control performance and operational complexities.…”