... | ... | @@ -98,11 +98,11 @@ Due to the black surface being round and having a white edge, we could no longer |
|
|
|
|
|
*Figure 4: Using a box to angle the robot for calibration.*
|
|
|
|
|
|
A problem we discovered with this box was that in some places it was casting a shadow over the surface that we were calibrating the robot on, causing the initial calibration value to differ largely from what it would be once the box was removed. We therefore made a conscious effort to always place the box and robot in a position where the box wasn't shadowing the robot.
|
|
|
A problem discovered with this box was that in some positions it was casting a shadow over the surface that we were calibrating the robot on, causing the initial calibration value to differ largely from what it would be once the box was removed. We therefore made a conscious effort to always place the box and robot in a position where the box was not casting shade on the robot.
|
|
|
|
|
|
Our experiments with this setup once again did not show any noticeable improvements in the balancing ability of the robot, despite the new surface.
|
|
|
|
|
|
After all of these various experiments with change of different factors, we came to the conclusion that this program simply doesn't allow the robot to balance very well under any conditions, and out of all the variables we could change, the (by far) most significant factor was the calibration position/angle.
|
|
|
After all of these various experiments with change of different factors, we came to the conclusion that this program simply doesn't allow the robot to balance very well under any conditions available to us, and out of all the variables we could change, the most significant factor (by far) was the calibration position/angle.
|
|
|
|
|
|
|
|
|
#### Algorithm-discussion
|
... | ... | @@ -264,6 +264,8 @@ We would have attempted to amend the inaccurate angle calculation and then use t |
|
|
|
|
|
Firstly, we should note that we were far from completing our goal of finishing the exercises within five hours. We will take this into account in the following lessons, either by allotting more time or by planning the exercises in more detail and sticking to a stricter plan.
|
|
|
|
|
|
Using Hurbain's three suggestions for optimizing the conditions for the robot to act in resulted in one particular main insight that could be especially useful later, which is that the calibration angle has significant impact on the robot's ability to balance and must take into account the weight distribution of the robot's physical appearance.
|
|
|
|
|
|
Our experimentation with creating a balancing robot using the light sensor proved to be immensely difficult. Systematically attempting different values for our P, I, D and setpoint variables proved somewhat fruitful, as we managed to somewhat remedy some of the issues that appeared to be causing the robot to fall over, but in the end it was still unable to balance for more than a second or two at best. A big culprit leading to this was also the mysterious random stopping of the motors that we were unable to find the source of.
|
|
|
|
|
|
Testing the parameters in a way inspired by grid search provided us with a structured approach to investigating how the different paramters influence each other. Although limited time forced us to move on before fully completing an actual grid search, we did obtain some insights into the effect of the parameters on the robot's responsiveness to the light sensor.
|
... | ... | |