Computer Vision and Machine Learning Reveal Developmental Changes in Continuous Motor Planning

Perception and action comprise a continual loop, where the just-prior action generates perceptual information for planning the upcoming action. Building a block tower requires updates to motor planning in a chain of perception-action loops. Each new block requires updates in the trajectory of the acting hand to place the block and updates in the position of the supporting hand to hold the tower steady. Eight-six participants (four-, six-, and eight-year-old children and adults, n = 20-23 per age group) built towers in a stable, “flush” condition (all blocks aligned) and an unstable, “offset” condition (initial block cantilevered). Head-mounted eye tracking recorded point of gaze; computer vision tracked hand and block movements; and machine learning tested whether participants used perceptual information about tower height and stability to update their motor plans. With age, children built higher towers with fewer breaks and shifted gaze earlier to the new block. Eight-year-olds and adults systematically changed the position of their supporting hand based on tower height and whether the condition was flush or offset, but younger children did not. Importantly, the spatiotemporal patterns of movements in both acting and supporting hands were similar for four- and six-year-olds and for eight-year-olds and adults, but differed between younger and older age groups. Findings show that children learn to update continuous motor plans using perceptual information about changes in the environment caused by their own just-prior actions. Moreover, findings illustrate the benefits of computer vision and machine learning to reveal developmental changes in motor planning.