Collective Decision-Making and Change Detection with Bayesian Robots in Dynamic Environments

The following folder structure holds all research data of my conducted experiments(h5-logfiles and plots). The Python-Script "show_h5.py" can be used to read out the logfile in h5-format ($python3 show_h5.py expample_logfilename.h5). However, this shouldn't be necessary because all plots are already generated.

To find the results you want to see, this is a small guide through the structure:

1. First the trials are divided into the respective methods (PELT, DBB, DBBCPD). In the folders you find the experiments for the specific method.

2. In the folder of PELT you find the results for the different feedback types and their combinations. The id for each feedback is noted in parentheses (e.g. XX_(id)_feedback_description). Feedback combinations have their ids added up (e.g. XX_(id1+...+idn)_feedback_description).

3. In the folder to each feedback type the different test trials can be found. This means varying environment difficulties and parameter settings. In the name of the folders this information can be found (e.g. XX_method_environmentdifficulty_parametersetting).

All experiments follow the same procedure as long as it is stated otherwise. Each trial consists of 20 individual runs with a duration of 6000 seconds. At half time (3000 s) a change to the opposite fill ratio occurs (fill ratio of 1.0 defines a completely white and one of 0.0 a completely black environment).

Environment difficulty

• 0901 --> easy environment, fill ratio changed from 0.9 to 0.1

• 0703 --> easy environment, fill ratio changed from 0.7 to 0.3

• 0604 --> easy environment, fill ratio changed from 0.6 to 0.4

• 055045 --> easy environment, fill ratio changed from 0.55 to 0.45

Parameter Setting

The setting is in the name of the folder composed of: feedbackID: intervalLength amountNeighbors

• 3c:50s3n --> feedback 3c with a 50s interval and 3 neighbors

In these folders all plots of the respective runs can be found showing a Boxplot of all 20 runs and for each run the swarm belief, the decision distribution and the reset histogram (before/after the change)