| Resumo: | Genetic algorithms are inspired by the process of natural selection that exists in nature. This process is what leads species to evolve and adapt to their surroundings, with the fittest species reproducing, leading to new generations that can take advantage of their surroundings better than before. This type of process can be used in evolutionary robotics to achieve controllers that are able to solve specific tasks to evolve morphologies for a specific purpose such as to walk, swim, grasp objects, among others. Robotic grippers are used in most factories nowadays, as well as in other workplaces such as hospitals and laboratories. They are used in tasks such as grabbing/moving objects, painting, surgeries, among many other uses. Grippers are therefore a case study with several possibilities that lend themselves to evolving morphologies through genetic algorithms. In this dissertation, we explore morphology generation through genetic algorithms. Using grippers as our case study, we were able to generate grippers capable of grabbing and lifting an object. To evolve these grippers, we created a simulated environment where grippers followed a script with instructions to grab the object and then move up. In total 120 different grippers were generated in these experiments. Out of those 120 generated grippers, 28% were able to grab and lift an object successfully. After the evaluation process was completed, we experimented with the grippers in five different scenarios to test their robustness. In these scenarios, the object’s starting conditions were different from those in the evaluation process.
|
|---|