Search papers, labs, and topics across Lattice.
This study investigates the role of embodiment in the motor development of infants by employing MIMo, a virtual infant model that utilizes proprioception and vestibular sensation to learn rolling behavior through reinforcement learning. The findings reveal that MIMo's learned rolling behaviors reflect developmental trends observed in real infants, demonstrating improved performance and speed as the model matures. This research underscores the significance of body morphology in shaping motor capabilities, providing insights into how embodied models can enhance our understanding of sensorimotor development in artificial agents.
MIMo's rolling behavior mirrors real infant development, revealing that motor skills evolve with changing body morphology.
Rolling over is one of the earliest milestones in infant motor development, reflecting the emergence of coordinated, whole-body sensorimotor control. Here, we conduct a computational study of infant rolling using MIMo, a virtual infant embodiment equipped with proprioception and vestibular sensation. MIMo learns supine-to-prone rolls with reinforcement learning. Interestingly, the learned behaviors capture developmental trends and coordination patterns consistent with those reported in real infants, including improved performance and faster execution with age. Our results explain how infant capabilities and constraints can give rise to realistic behaviors in artificial agents, with a particular emphasis on how motor development is shaped by the changing body morphology. This work highlights the role of embodied computational models as a powerful tool for studying sensorimotor development.
