*5.5. Discussion*

In the following, we first address the research questions defined in Section 1 and then we discuss our findings.

TuttiFrutti selects, tunes and assembles control software that operates with information that is available in the environment in the form of colors. In the three missions, the robot swarm reacts to these colors and act according to the information they provide in each case—both for handling events and navigating. Additionally, we observed that TuttiFrutti can design collective behaviors that exhibit color-based communication between robots. For example, TuttiFrutti designed collective behaviors with color-based communication in STOP and AGGREGATION—missions in which communication can influence the performance of the robot swarm. These collective behaviors are feasible thanks to the extended capabilities of the e-puck, capabilities that translated into a larger space of possible control software than the one considered by Vanilla and Chocolate—early versions of AutoMoDe. As the design space of TuttiFrutti is larger than the one of Vanilla and Chocolate, one could have expected that the automatic design process would have difficulties in producing meaningful control software. Still, we did not find evidence that TuttiFrutti suffers from an increased difficulty to design collective behaviors for robot swarms. The reference model RM 3 and the set of modules introduced with TuttiFrutti allowed it to conceive STOP and AGGREGATION—variants of missions already studied with AutoMoDe, and FORAGING—a new mission framed within the *best-of-n* problem. By introducing TuttiFrutti, we enlarged the variety of collective behaviors that can be produced with the AutoMoDe family.

We argue that the experiments we conducted with TuttiFrutti show evidence that automatic modular design methods can establish a mission-specific relationship between the colors that the robots perceive and the behavior that they must adopt. In Section 2, we described experiments in which this relationship enabled the design of complex collective behaviors [25,31,35]. We find that these collective behaviors have similarities with those designed by TuttiFrutti—for example, robots react to colored objects in the environment and use colors signals to communicate with their peers. We conjecture that TuttiFrutti, or design methods that might share its characteristics, can produce a wider range and more complex collective behaviors than those described in this paper. In this sense, we believe that research with robot swarms that can perceive and display colors has the potential to close the gap between the complexity of the missions performed with manual design methods, and those performed with automatic design.
