Model Predictive Control for Autonomous Micro Aerial Vehicles

Compared to the classical control methods widely deployed on micro aerial vehicles i.e. quadcopters/octocopters, such as PID etc., a series of advanced alternatives have recently been developed and applied on MAVs. In our lab, we implement one of the optimal control methods, model predictive control, on our quadrotor flying robots.

Micro Aerial Vehicle

We use a Mikrokopter QuadroXL quadrotor helicopter as a test platform, which is implemented with an Odroid XU board that contains a Cortex-A15 quad core CPU, 64GB SSD and a wireless network card. The quadcopter is mathematically modelled in 6DOF based on the kinetic and dynamic equations. In order to identify the MAV parameters and apply them on the controller, a 3D CAD model is also developed.

Fig. 1 The Micro Aerial Vehicle (MAV) platform and its CAD model.

Model Predictive Control

Model predictive control (MPC) is an advanced feedback control method that predicts the change in the dependent variables of the modelled system. The scheme minimizes a cost function by computing a sequence of optimal control inputs for a UAV model with constraints. Only the first control input of the sequence is then implemented. At the next time step, the calculations are repeated starting from the current state, obtaining a new control input as well as a new predicted trajectory. This is also known as receding horizon control. The following block diagram illustrates the whole picture of our control strategy.

Fig. 2 Block diagram of MPC for quadrotor helicopters.

Results

In flight tests, a 20Hz controller is executed in environments with artificial stochastic noise, in order to test the controller robustness. Results of the case that our quadcopter is controlled to autonomously follow the pre-defined rectangular trajectory show the performance of our MAV control algorithm.

Fig. 3 Comparison of quadcopter flight trajectory and pre-defined rectangular path in simulation (Matlab).