Obstacle-Aware Game-Theoretic Model Predictive Control for Quadrotor Pursuit-Evasion

Abstract

Effective autonomous pursuit of an intelligent adversary in complex environments requires more than reactive tracking or passive target extrapolation. This study introduces OA-GT-MPC, an obstacle-aware game-theoretic model predictive control framework for quadrotor pursuit under local perception. Unlike prediction-only interception methods, OA-GT-MPC integrates the evader's obstacle-constrained best-response optimization within the pursuer's receding-horizon interception problem. This integration ensures that the predicted evader trajectory is action-coupled to the candidate pursuer action, rather than being generated by an external extrapolator. To ensure safety under decentralized perception, a constraint-tightening-based conditional one-cycle hard-clearance preservation and a conditional recursive-feasibility result are derived. Simulations and onboard experiments in cluttered environments demonstrate both interception performance and real-time feasibility. OA-GT-MPC reduces capture time by 46.7% compared to Standard MPC baselines and 28% compared to the prediction-only baseline, with no observed obstacle collisions.