Robot-assisted rehabilitation which promises to redefine clinical strategies has drawn more and more attention for stroke therapy recently. Compared with classic rehabilitation limited to subjective observation of therapists on patients, the treatment by robotic devices can be more precise to improve the rehabilitation effects. In this paper, a wearable and portable rehabilitation robot with upper limbs of 5 degrees of freedom (DOFs) and upper limbs of 4 DOFs was introduced. The surface electromyographic (sEMG) signals were used to drive all the DOFs of both limbs to aid the training of prosthetic limbs. All data during the rehabilitation, including the sEMG signals and joint motions, can be collected for the purposes of both online analysis and offline storage.