Supplemental material: Comparing approaches for evolving high-level robot control based on behaviour repertoires

<p>Supplemental material for the paper:</p> <p>Comparing approaches for evolving high-level robot control based on behaviour repertoires<br> Jorge Gomes, Anders Lyhne Christensen</p> <p>Videos of the highest-performing solutions evolved with each method. In all videos, the schematic&nbsp;of the robot indicates the current wheel angles and robot speed. In the repertoire-based methods, the visualization on the bottom left indicates which primitives are being executed.</p> <p><strong>vid_evorbc_2_4_f1.12.mp4&nbsp;</strong><br> Evolved by NEAT-EvoRBC. The blue cross indicates the behaviour selection vector.</p> <p><strong>vid_gp_2_2_f1.12_best.mp4</strong><br> Evolved by GP-DT. The primitives highlighted in green&nbsp;are the ones that exist in the decision tree.</p> <p><strong>vid_gp_8_4_f0.92_paper.mp4&nbsp;</strong><br> Evolved by GP-DT. Smallest tree with fitness above 0.9. Shown in the paper.</p> <p><strong>vid_neatsub_9_5_f0.89.mp4&nbsp;</strong><br> Evolved by NEAT-Subset. Only the primitives that are in the subset are shown. The numbers near each primitive indicate the activation level of the corresponding output (or maximum activation level if there are multiple outputs corresponding to the same primitive).</p> <p><strong>vid_neattr_24_f0.28.mp4&nbsp;</strong><br> Evolved by NEAT-TR.</p> <p>The videos have been verified to reproduce correctly with the <strong>VLC player</strong>. But any other modern player should handle them.</p>