This paper presents an innovative nonlinear model of a self-sensing soft actuator and a switching extended Kalman filter, which is used as an observer for health monitoring such as a virtual sensor. Hardware-in-the-loop simulations and measurements justify the observation strategy. An adaptive position control structure based on the discrete wavelet packets transform method is proposed. Stimulus-responsive polymers are relevant for the realization of smart systems (capable of both sensing and actuating) in the context of soft actuators, for example, in bioinspired robotics, orthotic, and prosthetic technology. A real-time implementation of health monitoring techniques is important to guarantee a gentle, fault-free operation of the soft actuator in the presence of loads. The presented case shows a microrobotic application and measured results show the effectiveness of the developed observer regarding the soft actuator temperature to guarantee its longevity. The latter is necessary for normal operation without a test bench, because the states cannot be measured, but they need to be known for the implementation of the proposed health monitoring algorithms.