´Emergence de Comportements Adaptatifs et ´Evolutifs de Robots Modulaires

Modular robots are robotic systems composed of several interconnected modules; each module is self-contained with its sensing, actuating, computing, and communication means; these simple modules are connected to form more complex robots. Self reconfiguration is a property in modular robots that enable them to change their morphology autonomously to suit a specific task. Self-reconfiguration provides modular robots with more versatility and flexibility as opposite to single-purpose robots, and they represent a significant step towards building the universal robot. The potentials of modular robots are enormous, and it can be the perfect candidate for applications where the task is not entirely known in advance and rough and changing environments like rescue and space missions. The design of a modular robotic system is a very challenging task, and several design elements have to be considered. Modular robotics systems can be classified into different categories, chain-type, lattice-type, hybrid, and mobile; each of these types has its advantages and drawbacks and can dictate later the type of functions the robot can accomplish. Our proposed modular robot falls into the hybrid category; furthermore, it also demonstrates self-mobile capabilities. The main design goal behind our module is creating a simpler version of the previous hybrid modular robots; nonetheless, it must be fully capable of replicating all their reconfiguration strategies. In this thesis, we have used quantum genetic algorithms (QGAs) combined with artificial neural networks to evolve suitable controllers for our modular robot. Quantum-inspired evolutionary algorithms represent a significant advancement over conventional evolutionary algorithms; it combines the probabilistic search methods of evolutionary algorithms with the concepts of quantum computing like superposition, measurement, and interference. We have experimented with several QGAs variants, and real-observation QGA achieved the best results in solving numerical optimization problems. The combination of our module design and our neural controllers evolved using QGAs was able to produce a modular robot capable of adaptive locomotion and self-reconfiguration