Constraint Learning in Multi-Agent Dynamic Games from Demonstrations of Local Nash Interactions
This work addresses the challenge of constraint learning in multi-agent systems for robotics and autonomous systems, representing an incremental advancement by applying existing inverse dynamic game methods to learn constraints from Nash equilibrium data.
The paper tackles the problem of learning parametric constraints from demonstrations of local Nash equilibrium interactions in multi-agent dynamic games, using a mixed-integer linear program based on KKT conditions to recover constraints consistent with the demonstrations. The result includes theoretical guarantees for learning inner approximations of safe and unsafe sets and successful application in simulations and hardware experiments for inferring constraints and designing interactive motion plans across convex and non-convex cases.
We present an inverse dynamic game-based algorithm to learn parametric constraints from a given dataset of local generalized Nash equilibrium interactions between multiple agents. Specifically, we introduce mixed-integer linear programs (MILP) encoding the Karush-Kuhn-Tucker (KKT) conditions of the interacting agents, which recover constraints consistent with the Nash stationarity of the interaction demonstrations. We establish theoretical guarantees that our method learns inner approximations of the true safe and unsafe sets, as well as limitations of constraint learnability from demonstrations of Nash equilibrium interactions. We also use the interaction constraints recovered by our method to design motion plans that robustly satisfy the underlying constraints. Across simulations and hardware experiments, our methods proved capable of inferring constraints and designing interactive motion plans for various classes of constraints, both convex and non-convex, from interaction demonstrations of agents with nonlinear dynamics.