Bioengineering of joint action: from interpersonal interaction to neurorehabilitation

Joint action in sensorimotor contexts provides a complex yet essential framework for understanding how individuals negotiate actions, coordinate efforts, and make strategic decisions under uncertainty. While classical Game Theory (GT) offers a rigorous theoretical foundation for identifying equilibria in social interactions, it mainly operates within discrete action spaces and cognitive domains. A significant gap remains in understanding how these abstract strategic behaviors—ranging from the coordination required in mutualistic tasks to the unpredictability necessary for competitive success—are physically embodied within continuous, high-dimensional motor tasks. This thesis addresses this gap by developing a unified computational and experimental framework that bridges the divide between formal game-theoretic predictions and real-world sensorimotor execution. First of all, this research establishes a foundational computational framework capable of modeling a broad spectrum of interactions, including coordination and competition. In order to apply/extend the results and predictions of game theory to sensorimotor interaction, a primary challenge is to define sensorimotor versions of classical games, where action costs reflect mechanical (or physical) effort. We propose a general methodology using a quadratic optimization framework and a redefinition of interaction stiffness to ensure the mathematical integrity of the Nash Equilibrium (NE) is preserved throughout the continuous workspace. This framework was empirically validated through a sensorimotor version of the Prisoner's Dilemma (PD), where experimental results confirmed that human dyads successfully converge to the predicted sensorimotor NE by modulating their physical compliance and effort. The framework was further extended to address the stochastic nature of human decision-making through the study of Mixed Strategy Nash Equilibria (MSNE). We simulated stochastic interaction in classical coordination paradigms (Battle of the Sexes and Matching Pennies). These simulations revealed that behavioral variability is not merely a byproduct of neuromotor noise, but rather an active, strategic deployment of uncertainty. Crucially, we found that developing stable interaction strategies in these games depends on a dynamic internal partner representation—a model that agents continuously update based on the perceived reliability and past actions of their opponent. Furthermore, through an embodied Penalty Kick paradigm, we isolated the effects imperfect information on strategic stability. The results show that sensory uncertainty fundamentally shifts the strategic landscape; under conditions of imperfect information, participants often retreat to conservative, sub-optimal motor behaviors to prioritize physical safety and risk mitigation. These studies may have implications in neurorehabilitation where patient engagement is a fundamental aspect during trainings. As a first application of sensorimotor joint action in neurorehabilitation, we evaluated, through a pilot study with chronic stroke survivors, the efficacy of human-human haptic interaction against traditional robotic trajectory guidance in an ankle training. The findings challenge the conventional robot-guidance approach, revealing that the perfect kinematic information provided by robots can lead to a slacking effect and reduced active engagement. In contrast, interaction with a human partner—characterized by natural variability—promoted superior engagement and short-term learning. In conclusion, by bridging the gap between different Game Theory paradigms and motor control, this work provides a comprehensive understanding of the mechanisms underlying embodied joint action. These results can help in developing bio-inspired controllers and artificial partners capable of maximizing patient engagement and promoting motor recovery in adaptive neurorehabilitation systems.