We measure scalability, not by an internal measure of self-assembling efficiency, but rather by the external measure of task completion efficiency.We conduct our experiments using a simulated version of the swarm-bots robotic platform.The platform consists of a number of mobile autonomous robots called s-bots(see Fig.1)that are capable of forming physical connections with each other.Each s-bot is equipped with an XScale CPU running at 400 MHz, a number of sensors including an infrared ground sensors, proximity sensors, and light sensors.Physical connections between s-bots are established by a gripper-based connection mechanism.Each s-bot is surrounded by a semi-transparent ring that can be grasped by other s-bots.S-bots can advertise their location and/or internal state by means of eight sets of RGB-colored LEDs distributed around the inside of their semi-transparent ring.The s-bots have an omni-directional camera that points upwards at a hemispherical mirror mounted above the sbot’s turret in a transparent perspex tube.The camera records the panoramic images reflected in the mirror.Depending on light conditions, the camera can detect illuminated LEDs on other s-bots up to 50 cm away.The combination of the camera and the LEDs thus provides the s-bots with local, situated communication capabilities.The experiments in this study were conducted in a simulation environment consisting of a specialized software simulator with a custom dynamics engine tailored to our robotic platform[4].All the sensors and actuators that were used are simulated with reasonable accuracy by our simulation environment.We developed a control interface abstraction layer that allowed us to transfer our control programs between the simulator and the real robots without any modification.The control abstraction layer allowed us to run and test the same SWARMORPH-based control programs both in simulation and on real robots.TASKS AND MORPHOLOGIESWe have chosen three tasks:gap crossing, bridge traversal, and object pushing.None of these tasks can be solved by a single robot operating alone.Instead, the robots have to self-assemble and cooperate in order to accomplish each of the three tasks.Based on trial and error experimentation with real robots, we have designed the three tasks so that each task requires the robots to self-assemble into a dedicated morphology.Each morphology can solve one task and one task only, that is, the dedicated morphology that succeeds in solving one of the tasks will fail to solve if applied to either of the other two tasks.