This article is concerned with achieving resilient consensus for multi-agent systems under cyber attacks. Without loss of generality, we consider a leader-following multi-agent system where each agent is modeled with a Lipschitz nonlinear model, and the actuator attacks are imposed on agents’ states. To begin with, an adaptive observer is constructed to estimate the attacks for each follower based on relative output errors with respect to its neighbors and the leader. The global error system is proved to be asymptotically stable. Moreover, a resilient consensus control protocol is proposed for each follower where the guaranteed-performance is considered. Under this control protocol, the impact of the attack is compensated and the state of each agent is consistent. Different from the existing secure control scheme, the new one can simultaneously resist multiple types of attacks. Finally, simulation results are given to verify the effectiveness of attack estimation and resilient consensus control protocol for multi–unmanned aerial vehicle systems.