how to get the exact thrust f_u about neural-fly HOT 3 CLOSED

The thrust f_u is measured using a thrust stand to get an estimate of the installed thrust for each motor. The difference (or residual) between the observed thrust (i.e., what you have as the 'exact thrust') and the estimate is then learnt via the neural network. In our particular implementation (due to not being able to directly measure the RPM). we just use the thrust command, which is a throttle value between 0 and 1. Then we re-scale it by (m*g/) to convert to Newtons.

from neural-fly.

The thrust f_u is measured using a thrust stand to get an estimate of the installed thrust for each motor. The difference (or residual) between the observed thrust (i.e., what you have as the 'exact thrust') and the estimate is then learnt via the neural network. In our particular implementation (due to not being able to directly measure the RPM). we just use the thrust command, which is a throttle value between 0 and 1. Then we re-scale it by (m*g/) to convert to Newtons.

Thanks for your reply.
And In your paper it is mention that you compute the acceleration by fifth-order numerical differentiation, why you don't use the acceleration values of imu.

from neural-fly.

We don't receive the full rate IMU data, so numerical differentiation of the EKF data was better for our system. It may have been better to use the IMU data if we had full rate and could low pass filter it. Also, for adaptive control we used one step backwards Euler difference method. The 5 step central difference was only for training. Either way, these are more implementation details and not necessarily the best way for every system.

from neural-fly.