Robotic Invention: Challenges and Perspectives for Model-Free Design Optimization of Dynamic Locomotion Robots

Abstract

To improve a robot’s performance at a given task, or to respond to changing requirements, shape adaptation can be beneficial. To efficiently explore complex behaviors, diverse morphologies must be generated and implemented. For continuous and autonomous design optimization, the robot has furthermore to be able to assess its own performance and in turn generate and implement adapted morphological designs. Here, we present the morphological adaptation of physical robotic agents to a locomotion task. The robots are automatically assembled by a robotic manipulator from elementary modules and the assembly process of each agent is encoded in a genotype. The genotypes of a robot population are optimized using an evolutionary algorithm based on real-world performance feedback. In the experiments, 500 genotypes were evaluated. To develop rich behavioral diversity, shape variations are beneficial. Analysis of the results highlights the influence of the fabrication constraints on shape diversity, which impose limitations especially for larger structures.