Optimization of Mass Distribution in Articulated Figures with Genetic Algorithms

摘要

The movement of the human body is simulated with a virtual, mechanical, robot-type model (articulated figure) consisting of 14 solid subbodies. The shapes of these 14 subbodies are initially a ball, cylinders or rectangular boxes. The model is initially adapted, in particular the dimensions of the body, to fit measured data taken from a human athlete’s high jump.The initial model’s subbodies are then allowed to vary in shape, but not in mass. For a particular shape of the articulated figure, the mechanical energy or work consumed during the high jump (following the measured trajectories of the human athlete) is computed applying the Euler-Lagrange formalism of classical mechanics. This energy value is set to the inverse fitness value of the corresponding shape in an otherwise custom designed genetic algorithm, where the shape of the articulated figure represents an individual or creature in the population.Under a simple continuity assumption for the shape of the 14 solid subbodies of the articulated figure, we obtain an optimized figure which shows interesting features such as “cone-shaped muscle groups” or “hips.” These features are stable under repeated application of the genetic algorithm procedure.