The authors present an adaptive optimal control algorithm for uncertain nonlinear systems. A variational technique based on Pontryagin's maximum principle is used to track the system's unknown terms and to calculate the optimal control. The reformulation of the variational technique as an initial value problem allows this microprocessor-based algorithm to perform as a closed-loop controller by updating the model and controlling the system online. To validate the algorithm a system representing a two-link mechanical manipulator is simulated. In the control model, the coupling and friction terms are unknown. The robot's task is to follow a prescribed trajectory and to pick up an unknown mass. >