SCB #1 2: Autonomous Flight - maxct/paparazzi GitHub Wiki

Introduction

Second scenario is Autonomous Flight. In this scenario we investigate behavior of single copter from the beginning of "inair" part of flight plan to the manual landing after 60 (120 for Finken 3_02) seconds. In future, we need to know current behavior of copters to achieve at least the same stable behavior with to autonomous copters. Also we need to know, how real behavior is different from idealized behavior. We need this mistake to check, how second copter affect behavior of first during simultaneous flight.

Description of experiment

We've got data for this scenario by launching both copters for several times in wall avoidance mode. We set time of default part equal to 10 seconds. Moreover, every experiment begins from manual switching to "default" part of flight plan, but recording of telemetry starts from launching server. Therefore, we need to cut all information received before this switch. This have been done manually for every experiment. During experiments we had to change calibration parameters to achieve stable behavior. But in some cases copter acts not very stable, so in underlying evaluations only data from stable flights are used. Data of experiment located in Google Drive.

Evaluating of data for single flight

In this section usual behavior of copter are described. For all calculations and plots data from third flight of Finken 3_02 are used. After one can see generalization of observed behavior and calculated values.

General description

In every flight for both copters there are some common events:

  • At first, copter jumps from floor to the approximate height of 0.4m and then height decreases to almost zero.
  • Then, during some time interval copter gradually increases it's height with some oscillations.
  • After that, during rest flight height of copter stays near 0.6m, as it should be. Some changes in height is caused by moving of copter.

In first plot height during overall flight is showed.

height

In next sections every step are described more precisely.

Height increasing

For this exact experiment one can observe:

  • height increases for 30 seconds;
  • Frequency of oscillations is approximately 1.45Hz.

Frequency was calculated by such procedure:

  • Height was smoothed by using [Gaussian filter](https://en.wikipedia.org/wiki/Gaussian_filter with radius equals to 30. );
  • Difference between original and smoothed signal was calculated;
  • Amount of sign's changing of difference was calculated. And then it was divided by number of observations.

Next plot shows this. Also, we can observe, that height increase is almost linear. start

Stable part of flight

After height stabilization copter stays in 0.56m. Increases and decreases of height are caused copter position changing because of wall avoidance. It is normal behavior. To compare differences we need to check the "error" of height. Error in this case is quadratic difference between original height and height with Gaussian filter applied. Radius still equals to 30. Result error equals to 0.00036.

Two next plots shows mistake filtered height with original and changing of mistake through time.

main

main_noise

##General Evaluations

Finken 3_02

  • Jump. 0.44, 0.3, 0.37. Height is rather different, but it can be explained by battery level.
  • Height increasing. 32, 40, 41 second. Numbers depends on what we set as end of height increasing. But most valued is almost equal behavior and time of stabilization.
  • Frequency of oscillations during stabilization. 1.46, 1.44. Also near the same in described experiments.
  • Average height after stabilization. 0.56, 0.55, 0.57. This parameter near the same in every experiment.
  • Overall error. 0.00038, 0.00052, 0.00044. We noticed in lab, that second flight was most unstable. Error confirms this. Absolute value of mistake can't tell us anything yet. We need data of simultaneous flights.

Finken 3_01

  • Jump. 0.42, 0.35, 0.41.
  • Height increasing. 28, 40, 33 second. There is tendency: bigger jump, faster height increase. But not every time.
  • Frequency of oscillations during stabilization. 1.68, 1.44, 1.69. Also near the same in described experiments.
  • Average height after stabilization. 0.56, 0.56, 0.57. This parameter near the same in every experiment.
  • Overall error. 0.00044, 0.00025, 0.00020.

Some other observations

In last plot we can see yaw changing during autonomous flight. In every flight we've noticed yaw approximately 180 degrees in one minute. And this plot shows us almost linear rotation.

yaw