Traditional rehabilitation suffers from aplenty downfalls; they are costly and time consuming. Robotic rehabilitation has the potential to be a better substitute. Recent evidence suggests that there is a pressing need to employ the patient muscle effort to control the assistive robot, otherwise the patient may fully depend on the robot, and this leads to slackness and deteriorated muscle functionalities. The development of a non-invasive human-machine interface is challenging, since surface electromyography (sEMG) electrodes are uncertain and noise; hence a model that considers the uncertainty and noise involved seems important. This study aims to contribute to this growing area of research by exploring and comparing the performance of different Fuzzy logic techniques on the estimation of joint torques from relevant muscles electromyography measurements for the accurate control of rehabilitation exoskeletons. © 2016 IEEE.