A robust adaptive compensation scheme is presented for compensation of asymmetric deadzone, dynamic friction and uncertainty in the direct-drive robot manipulator. Simple estimation laws are derived to build observers for estimation of deadzone and friction based on the LuGre friction model. A model-free RWCMAC controller to mimic the ideal control law is employed to overcome some shortcomings of the traditional model-based adaptive controller, which requires information on the robots dynamics in advance. The Lyapunov stability analysis yields the adaptive laws of the RWCMAC network as well as observers of deadzone and friction. Furthermore, the stability and optimal convergence speed of the learning rates of the RWCMAC is also guaranteed by employing the fully informed particle swarm (FIPS) algorithm. Robust tracking performance of the proposed control schemes is verified by simulations of direct-drive robots with deadzone in joint input torque, joint dynamic friction and uncertainty.