Treatment of the results
To make a sensible comparison between the three algorithms, it is of high importance that they are tested on the same conditions. The conditions here are the settings of the sliders and buttons
in NetLogo. A choice has to be made for the following buttons and sliders:
- width-window (slider)
- height-window (slider)
- dirt-intensity (slider)
- nr-of-dirt-clusters (slider)
- min_dirtdecrease (slider)
- hard_to_clean_active (switch)
Regarding the width and the height of the window, it is decided to evaluate the algorithms on the
following window sizes: 1 x 1, 1 x 3, 3 x 1, 4 x 2, 2 x 4 and 4 x 4 m. These particular numbers
are chosen since they represent square and narrow as well as wide rectangular windows. Due
to this, it is possible to conclude that a certain algorithm is better for a certain window size. Every
window size is tested five times, each test containing an alternative dirt distribution. Note that
each test the dirt distribution was the same for each motion-planning algorithm. This was done
by importing the settings with the Import Window button. The results for time, energy and water
consumptions are averaged over the five tests to get more accurate results.
The dirt distribution together with the hard to clean patches in particular, determine a significant part of the overall time it takes for the robot to clean the window. Therefore, to compare the motion-planning algorithms, it is important to make a well-funded choice for the values of dirt-intensity, nr-of-dirt-clusters, min_dirtdecrease and hard_to_clean_active. For each test case it is decided to keep the values for these sliders and buttons the same. Overall, most of the values are set to the maximum values, since ’hard’ cleaning conditions will result in larger differences in performance and reveal the specific beneficial characteristics of the algorithms more clearly.
The slider dirt-intensity is set to 100%. This value will not have a major effect on the performance of the different algorithms. The slider nr-of-dirt-clusters is set to 10. This slider will have a little more impact, since the dirt values of clusters are significantly higher and may require the algorithms (the reference algorithm in particular) to do extra wiping motions. Subsequently, the slider min_dirtdecrease is set to 1. There are two reasons for this particular choice. When the slider is moved to its minimum, all the algorithms will clean the window till almost every single patch is clean, resulting in less differences in performance indicators. When the slider is moved to its maximum, apart from the hard to clean patches also a lot of other dirty patches remain, which also results in less difference in performance of the three algorithms. Hence the value of 1 is chosen since it gives a significant difference in performance for each algorithm. At last, the switch hard_to_clean_active is turned on, since these hard to clean patches are crucial for the algorithms and will have the largest effect on the performance of the three algorithms. Furthermore, this creates a test environment that is more realistic.
Since the model calculates the time, energy used and water used during the cleaning job, a MATLAB script is made to convert this data to performance indicators such as cleaning speed (m2/hr), water consumption (L/hr) and mean energy consumption (W).