Viewer CPU vs GPU MAP-Elites vs EA MAP-Elites
Press “New run”.
bone % of voxels →
filled % of grid →
0 m– distance in 10 s · empty cells are blank
Distance (m in 10 s) by iteration
best robot average of the top 50% of filled cells
Cells improved per 10 iterations (new or better elites; the initial random batch not counted)
Shows the best robot. Click a cell to pick another.

        

          
          
          
          
          
          
          
        

        

          
Its CPPN, queried at 256×256 instead of only at the voxel centres. Outlined voxels made it into the robot.

          
          

        

        

          
Lineage, body and behaviour (behaviour from the run that set its score)

          

        

      

      

        
Species, families and behaviour groups

        

          None of these read the genome, so they work the same for every genome kind. A family is everything
          descended from one robot of the initial random batch. A species starts from one robot; a child stays in
          its parent's species while its body stays within the threshold of that species' founder, and otherwise joins
          or founds a daughter species, so species follow descent and split only where bodies change a lot.
          A behaviour group comes from simple rules on the body and the scored run.
        

        

          
          
        

        

          
Share of the filled cells by iteration (a Muller plot: a daughter species grows out of its parent's band). Hover for details, click to highlight.

          

            
            
          

        

        

      

      

        
Robots shown

        
Every robot the replay has shown in this run, newest first, and why it was picked. Click a row to show that cell again.

        

        

          

            
timeiterationwhyspeciesfamilybehaviourdistanceage (iterations)map cell