An adaptive second-order sliding mode output feedback controller is developed to deal with the case that the bound of the uncertainty/perturbation is unknown. The control structure consists in a twisting controller and a super-twisting observer to estimate the unmeasured variable. The gains of the controller and observer are parametrized in terms of a scalar gain, such that increasing these two gains, it is always possible to find values to (finite-time) stabilize the closed loop system. Finally, adaptive gain laws are provided to increase the controller and the observer gains until the closed loop has been stabilized. The main technical contribution of the paper is to give a sound and non-trivial Lyapunov analysis of this otherwise intuitively simple idea. We illustrate the performance of the proposed controller by means of experimental results in a laboratory setup.Successful implementation of high-order sliding mode controllers requires the knowledge of the bound of the perturbation, because the appropriate gains for controller and differentiator are functions of these bounds [6][7][8][9]11]. These bounds are usually constants, but they can take also the form of a (known) time-varying signal [11]. When a bound for the perturbation is not known a priori, one is usually led to select an overestimated value, with the consequence of a high chattering amplitude. Recently, some strategies to adapt the values of the gains have been also proposed. For example, for classical sliding modes, Oliveira et al. [12] propose an adaptive approach to overcome limitations of uncalibrated robotics visual servoing. In [13], an adaptive second-order sliding mode controller is presented based on the estimation of the equivalent control by using only output measurements.Recently, for SOSM, two approaches to adapt the gain have been proposed. The first attack is to reconstruct the perturbation and to adapt the gain according to its estimated value. In [14], an adaptive super-twisting controller is presented. The adaptive gain 'follows' the equivalent control, which is obtained by filtering out the control signal by a low-pass filter. This method can track the required values of the gains very closely, but it requires the knowledge of the bound of the derivative of the perturbation. Furthermore, a delay on the control signal is presented because of the filtering process.The second approach to adapt the sliding mode controller gain is to detect the sliding mode: the idea is to increase the gain if the sliding mode is not established and decrease the gain when the sliding mode has been established. For example, in [15], an adaptive twisting controller is presented, assuming that the full state vector is available, while in [16], an adaptive super-twisting controller is developed. In [17], an adaptive super-twisting differentiator is designed for fault detection. In [18], an adaptive twisting controller is used on the joint position tracking control of industrial robots. The adaptation is based on detection of the sliding mode and is use...