In this paper, a novel adaptive robust controller is presented for high-accuracy tracking control of servomotor rigid robotic systems based on a modified nonsingular terminal sliding scheme incorporated with an adaptive time-delay estimation (TDE) technique. To effectively estimate unknown systematic dynamics, the model-free estimation is designed with a basic TDE and a new self-learning mechanism, which is able to automatically find the proper nominal input gains for the best TDE error. The control objective is then realized by a control process which consists of a model-compensation phase, highly robust control term and reaching-guarantee law. The compensator uses the TDE terms obtained to eliminate the dynamic effect. For providing the flexible robustness to the control system, the robust gain is actively adjusted based on information of the estimation system. The reaching-control rule employs nonlinear folding functions to promptly drive a synthesized sliding manifold back to zero. Finite-time stabilities of both the adaptive TDE and the closed loop system are proven using Lyapunov-based constraints extended from new folding-convergence analyses. The effectiveness and feasibility of the closed loop system are confirmed throughout intensive real-time experiments under various testing conditions.