Romero's Odyssey to Santa Fe: from Simulation to Real Life

Many a researcher in evolutionary robotics would love to have at his or her disposal a real population of bopping critters. Toward this end, the ultimate goal of our work is twofold: (1) to build a robot that is cheap, modular, and adaptable; and (2) to use a population of such robots in evolutionary-robotics experiments. As a rst step in the study of adaptive robotics using populations of cheap robots, we have designed the Lausanne trail, a realworld version of the simulated Santa Fe trail. We use Lego-based robots equipped with low-cost sensors and actuators to evolve solutions in the real world to the trail-following problem. We describe two di erent sets of experiments and conclude that the real-world problem exhibits fundamental di erences from the virtual one; in particular, the robot's lack of positional information hinders greatly its performance. We propose to study in the future collective map building, an important and interesting problem, which will, among others, provide a solution to the Lausanne-trail problem.