Objective Misspecification in Model Predictive Game Controllers: Stability and Sensitivity Analysis

Model-based multi-agent control requires each agent to consider or predict the behavior of others when making decisions. Game-theoretic solution concepts are often used to model the resulting strategic interactions and have been embedded within control architectures such as model predictive game (MPG) controllers. Within each MPG controller, an agent iteratively solves a finite-horizon game to predict future system behavior and compute its control action.
In practice, agents may hold inaccurate models of others’ objectives, leading to objective misspecifications among MPG controllers. This raises a fundamental question: how do inconsistencies between agents’ internal game models affect closed-loop behavior?
In this talk, we study the impact of such misspecifications on (i) closed-loop stability and (ii) the sensitivity of equilibrium outcomes to agents’ game parameters. Our results provide conditions under which stability can still be preserved despite model misspecification, and offer insight into how increasing inconsistencies between agents’ models impact the system equilibrium.

Join at imt.lu/aula2