Fig 1.
The robotic swarm with 10 units performing a homing task.
Fig 2.
The control synthesis and performance assessment process for each task.
Fig 3.
The robot is an autonomous surface vehicle equipped with Wi-Fi for communication, and a compass and GPS for navigation.
It has a length of 60 cm and can move at speeds of up to 1.7 m/s.
Fig 4.
Illustration of the three types of sensors.
Fig 5.
Fitness plot for the four different tasks.
The plot shows the highest fitness scores found so far at each generation. The red lines depict the three highest-scoring evolutionary runs, while the blue line depicts the average of the ten runs, with the respective standard deviation shown in gray.
Fig 6.
Panoramic photo of the location the experiments at Parque das Nações, Lisbon, Portugal.
Fig 7.
Real-world homing experiments with eight robots.
The robots started around S. The active waypoint was then changed at 60 second intervals, in the order A→B→C→B, for a total of four minutes per experiment. Top: comparison between the real and simulated robots, showing the average distance to the active waypoint, for similar conditions. The top of the figure shows the current active waypoint. Bottom: trajectories of the real robots for Controller 3. The waypoints are marked with yellow circles.
Fig 8.
Real-world dispersion experiments with eight robots, one for each controller tested, over a period of 90 seconds.
Top: average error to target distance (20 m) of the nearest robot in the last 10 s of each dispersion experiment. Bottom: trajectories of the real robots. The black squares mark the starting positions, and the red circles mark the final positions.
Fig 9.
Real-world clustering experiments with eight robots, over a period of 180 seconds.
Top: minimum number of clusters obtained in each sample. Bottom: trajectories of the real robots. The final clusters are highlighted in blue.
Fig 10.
Real-world monitoring experiments with eight robots for Controller 1, over a period of five minutes.
Top: coverage of the three different monitoring areas. Bottom: coverage maps in the experiments with the real swarm. The coverage of the area is presented in blue, and has a decay of 100 s. Trajectories for the full duration of the task are presented in red, and all the areas visited by the robots are filled in gray.
Fig 11.
Scalability experiments with dispersion (Controller 3, left) and clustering (Controller 1, right) controllers.
In each task, the same controller was used in a swarm of four, six, and eight robots, with three samples for each setup.
Fig 12.
Robustness experiments with Controller 3 of the dispersion behavior.
The red area represents the period where the robots of Gb are disturbing the dispersion of Ga, and the black vertical line at t = 180 s indicates the point where the robots in Gb start dispersing, and where the distance error starts being measured for all eight robots.
Fig 13.
Robustness experiments with Controller 1 of the monitoring behavior.
The time regions highlighted in red correspond to the periods when robots where either entering or leaving the monitoring area.
Fig 14.
Results for the multi-controller mission.
Top: robot trajectories for each sub-task. Middle and bottom: temperatures in the monitoring area. Measurements taken by the robots’ temperature sensors were spatially interpolated using kriging [71]. Data collection started after the robots arrived at the waypoint (t = 100 s). The middle row shows the predicted temperatures, while the bottom row shows the estimated error of the predictions.