This paper evaluates experimentally a novel strategy for solving a variant of the differential game of target defense in presence of obstacles. The game is widely applied in the areas of military defense for protecting important equipment such as a ship, an aircraft, a moving vehicle, or a sensitive installation from a malicious attacker. The state‐of‐the‐art approaches mostly employ an offline optimization strategy that is only applicable to holonomic robots. Moreover, most of the approaches could not autonomously avoid obstacles or take into account uncertainties. As a consequence, this paper presents an online optimization technique, by designing a trade‐off parameter that integrates game theory with the model predictive control, which allows a nonholonomic defender to intercept the attacker while simultaneously defending the target. Simulations under different conditions as well as several indoor laboratory experiments validate the proposed approach. Moreover, performance is compared with a standard model predictive control approach.