This master thesis details the design, development, and control of pleated cylindrical artificial muscles (PAM) for upper limb rehabilitation training. Recognizing the challenges associated with controlling complex PAM dynamics using traditional model-based approaches, this work proposes an intelligent adaptive controller based on input-output data for accurate trajectory tracking. Qualitative analysis guarantees the closed-loop stability, while quantitative evaluations assess trajectory error bounds. Extensive testing encompasses both computer simulations utilizing the developed PAM dynamic model and experimental validation on a prototype. Experimental results corroborate theoretical predictions and simulation findings, demonstrating superior performance compared to non-pleated AMs. Notably, PAMs exhibit reduced actuation dead zones and significantly enhanced actuation speed, making them attractive candidates for rehabilitation applications.