The role of accelerometers in the control of fully-actuated UAVs

MSc assignment

Conventional rotary-wing UAVs such as quadrotors and hexarotors are underactuated systems where typically 4-DOFs, the position (x,y,z) and heading (yaw angle) are the controlled DOFs. The collinear configuration of the propellers generate a total thrust vector along the normal to the horizontal plane of the vehicle, and the torques are generated by varying the speeds of the propellers. Recently, new designs of fully-actuated UAVs emerged, these have a non-collinear propellers configurations which can produce a full 6D wrench to control the 6DOFs independently.

As UAVs rely on inexpensive MEMS inertial measurement units (IMU) which has accelerometers and gyroscopes for state estimation and control, the measurements from these sensors are fused together using different kinds of filters (EKF, UKF, non-linear complementary filters, etc.) to estimate the attitude.

An accelerometer measures the specific acceleration, i.e., the difference between the acceleration and gravity, rotated in body frame. Typical filtering techniques make a coarse approximation that the vehicle acceleration is zero and therefore the accelerometer measures only the gravity vector, which is used as a low-frequency information source to reconstruct the roll and pitch angles (inclinometer). Furthermore, typical controllers are designed using a model that results in accelerometer measurements that would clearly preclude its use as inclinometer (because of its constantly zero lateral components). However closing the loop to control the vehicle attitude it is successful even in the presence of such manifest design flows that assume to use the accelerometer as an inclinometer. The authors in *[1]* provided an explanation describing the role of the accelerometer in the control, by including in the dynamic model of the quadrotor the aerodynamic effects that are function of propeller speed times the vehicle velocity (linear or angular).
The main significant quantity is the so called rotor-drag or (H-force as in helicopter theory). Their work triggered more people to consider the rotor-drag force to design better state-estimation and control of quadrotors.

This assignment is concerned about the control of a generic fully-actuated UAV that relies only on an IMU (accelerometers, gyroscopes and possibly a magnetometer), where there are no other sensing modalities for position, velocity and attitude (motion capture systems, GPS, visual odometry etc. ). To some extent it has been possible in our labs to use same UKF that is used for quadrotors to fly manually a fully-actuated UAV in two modes:
-underactuated : commanding the roll/pitch, yaw rate and the total thrust vector along the z-axis of the vehicle, (the vehicle has to tilt to translate laterally)
-fully-actuated: exploiting the vehicle ability to generate thrust (in other words accelerate in open loop fashion) in lateral directions without the need to change the attitude.

In the case of underactuated the lateral acceleration and attitude are coupled, which is not the case for fully-actuated UAVs. One of the main scientific questions of the thesis is to understand the role of the accelerometer in the case of fully-actuated UAV by mainly incorporating the rotor-drag in the model, and understanding the limits of standard attitude estimation techniques in controlling the UAV in both modes. With a potential of developing a new control/estimation technique for fully-actuated UAVs.

For validation and testing a newly built fully-actuated platform will be used (FiberTHex), along with a motion capture system (optitrack).

*[1]* Philippe Martin and Erwan Salaün, The true role of accelerometer feedback in quadrotor control, 2010 IEEE International Conference on Robotics and Automation (ICRA), pp. 1623--1629, Anchorage, AK, USA, DOI: 10.1109/ROBOT.2010.5509980, https://ieeexplore.ieee.org/document/5509980