Abstract
To harness the complexity of their high-dimensional bodies duringsensorimotor development, infants are guided by patterns of freezing andfreeing of degrees of freedom. For instance, when learning to reach, infantsfree the degrees of freedom in their arm proximodistally, i.e. from joints thatare closer to the body to those that are more distant. Here, we formulate andstudy computationally the hypothesis that such patterns can emergespontaneously as the result of a family of stochastic optimization processes(evolution strategies with covariance-matrix adaptation), without an innateencoding of a maturational schedule. In particular, we present simulatedexperiments with an arm where a computational learner progressively acquiresreaching skills through adaptive exploration, and we show that a proximodistalorganization appears spontaneously, which we denote PDFF (ProximoDistalFreezing and Freeing of degrees of freedom). We also compare this emergentorganization between different arm morphologies -- from human-like to quiteunnatural ones -- to study the effect of different kinematic structures on theemergence of PDFF. Keywords: human motor learning; proximo-distal exploration;stochastic optimization; modelling; evolution strategies; cross-entropymethods; policy search; morphology.}