This thesis is concerned with the design and analysis of behaviors in swarm robotic systems using minimal information acquisition and processing. The motivation for this work is to contribute in paving the way for the implementation of swarm robotic systems at physically small scales, which will open up new application domains for their operation. At these scales, the space and energy available for the integration of sensors and computational hardware within the individual robots is at a premium. As a result, trade-offs in performance can be justified if a task can be achieved in a more parsimonious way. A framework is developed whereby meaningful collective behaviors in swarms of robots can be shown to emerge without the robots, in principle, possessing any run-time memory or performing any arithmetic computations. This is achieved by the robots having only discrete-valued sensors, and purely reactive controllers. Black-box search methods are used to automatically synthesize these controllers for desired collective behaviors. This framework is successfully applied to two canonical tasks in swarm robotics: self-organized aggregation of robots, and self-organized clustering of objects by robots. In the case of aggregation, the robots are equipped with one binary sensor, which informs them whether or not there is another robot in their line of sight. This makes the structure of the robots’ controller simple enough that its entire space can be systematically searched to locate the optimal controller (within a finite resolution). In the case of object clustering, the robots’ sensor is extended to have three states, distinguishing between robots, objects, and the background. This still requires no run-time memory or arithmetic computations on the part of the robots. It is statistically shown that the extension of the sensor to have three states leads to a better performance as compared to the cases where the sensor is binary, and cannot distinguish between robots and objects, or robots and the background.