Early behavioral development as the structuring of sensorimotor information

Human fetuses and young infants behave spontaneously before the emergence of explicit goals and rewards. These behaviors, though appearing random at first glance, give rise to regulated sensorimotor interactions through the developing body and its environment, thereby contributing to neural circuit maturation and later development. In this chapter, we propose a framework that treats early development as the progressive structuring of sensorimotor information. We call this emerging regularity the sensorimotor information structure (SMIS), defined as the dynamic pattern of information flow among motor output and sensory input such as muscle activity, proprioception, touch, and vision. We review evidence that spontaneous activity and activity-dependent plasticity play central roles in early development and that embodiment provides strong constraints that organize these signals. We then present an empirical approach to early spontaneous movements that combines quantitative analysis with a biologically grounded musculoskeletal model. This approach enables the estimation of whole-body muscle activity and proprioceptive signals, revealing modular sensorimotor organization and recurrent state transitions during spontaneous movements. These SMIS patterns exhibit developmental changes over the first three months after birth, reflecting spontaneous exploration, which we term sensorimotor wandering. To move from description to mechanism, we introduce a constructive approach based on an embodied simulation that integrates a musculoskeletal model, multimodal sensory models, and spontaneous spinal activity, enabling key aspects of SMIS to be reproduced and their roles to be examined. Finally, we propose that early SMIS can serve as a developmental prior for later learning and provide a computational bridge from spontaneous movement to the emergence of agency and goal-directed behavior.