BUILDING NOVEL ACTION-OUTCOME MAPPINGS FOR SEQUENTIAL MOTOR SKILLS

Abstract

Understanding how humans acquire novel motor skills is a central topic in motor learning research. However, much of the work in this field has focused on adaptation experiments, leaving other key aspects of de novo skill acquisition less explored. For many de novo skills, individuals must learn new associations between discrete actions and arbitrary outcomes. This is evident in digital devices like video games, where pressing buttons on a controller can make a character jump or run. These action-outcome mappings are fundamental to the formation of the new skill. Therefore, understanding how they are learned and consolidated is essential for advancing our knowledge of motor skill acquisition and its application to various domains, from gaming to real-world tool use.In Chapter 2, using a task of grid navigation, I study how these action-outcome mappings are acquired and examine the role of training variability in the formation of generalizable mappings. Crucially, when a novel mapping is being learned, it often occurs within the context of sequential decision-making, allowing the interaction of motor learning and planning. In Chapter 3, I investigate this interaction with the aim of bridging the gap between motor sequence learning and planning research. Finally, in Chapter 4, I study the effectiveness of external contextual cues in the learning of multiple mappings, which have proven unsuccessful in standard motor adaptation experiments. The behavioral results from each chapter of this dissertation are complemented by computational models that integrate algorithms from reinforcement learning, tree search, and Bayesian learning. These models aim to provide insights into the cognitive processes underlying participants’ performance.