multi-agent systems; actuator saturation; three-layer neural networks; formation control; finite-time observer