An approach to the design of high performance sliding mode controllers for robot manipulators is presented. It employs a Takagi-Sugeno fuzzy system to describe the sliding surface. Each rule of this system represents the maximum slope sliding line for a certain set of parameters given in the premise part. Hence, the slope of the surface is adapted according to the current state of the manipulator. This new algorithm provides nearly time-optimal performance and still retains the robustness, typical for systems in sliding mode. The maximum slope sliding surfaces are designed using knowledge about robot's physical properties.