Objective: To evaluate if bilateral or unilateral upper limb robot-assisted rehabilitation training using a new three-dimensional end-effector robot that targets shoulder and elbow flexion and abduction is superior to conventional therapy with regard to upper extremity motor function recovery and neuromuscular improvement in stroke patients. Design: Randomized, controlled, parallel, assessor-blinded, three-arm clinical trial. Setting: Southeast University Zhongda Hospital Nanjing, Jiangsu, China. Methods: Seventy patients with hemiplegic stroke were randomly assigned to conventional training (Control, n = 23) or unilateral (URT, n = 23), or bilateral robotic training (BRT, n = 24). The conventional group received routine rehabilitation, 60 min/day, 6 days/week, for 3 weeks. For URT and BRT upper limb robot-assisted rehabilitation training was added. This was 60 min/day, 6 days/week, for 3 weeks. The primary outcome was upper limb motor function assessed with Fugl-Meyer–Upper Extremity Scale (FMA–UE). Secondary outcomes were activities of daily living (ADL) assessed with the Modified Barthel Index (MBI), Motor Evoked Potential (MEP) to assess corticospinal tract connectivity, Root Mean Square (RMS) value, and integrate Electromyography (iEMG) value recorded by surface electromyography to evaluate muscle contraction function. Results: The primary outcome indicator FMA–UE (least square mean (LSMEAN): 31.40, 95% confidence interval (95% CI): 27.74–35.07) and the secondary outcome indicator MBI (LSMEAN: 69.95, 95% CI: 66.69–73.21) were significantly improved in BRT as opposed to control (FMA–UE, LSMEAN: 24.79, 95% CI: 22.23–27.35; MBI, LSMEAN: 62.75, 95% CI: 59.42–66.09); and unilateral (FMA–UE, LSMEAN: 25.97, 95% CI: 23.57–28.36; MBI, LSMEAN: 64.34, 95% CI: 61.01–67.68). BRT also showed greater improvement in the anterior deltoid bundle with regard to muscle contraction function indicated by RMS (LSMEAN: 257.79, 95% CI: 211.45–304.12) and iEMG (LSMEAN: 202.01, 95% CI: 167.09–236.94), as compared to the controls (RMS, LSMEAN: 170.77, 95% CI: 148.97–192.58; iEMG, LSMEAN: 132.09, 95% CI: 114.51–149.68), and URT (RMS, LSMEAN: 179.05, 95% CI: 156.03–202.07; iEMG, LSMEAN: 130.38, 95% CI: 107.50–153.26). There was no statistically significant difference between URT and conventional training for any outcome. There was no significant difference in MEP extraction rate after treatment between groups (p = 0.54 for URT, p = 0.08 for BRT). Conclusions: A 60 min daily training for upper extremities using a three-dimensional end-effector targeting elbow and shoulder adding conventional rehabilitation appears to promote upper limb function and ADL in stroke patients only if delivered bilaterally. URT does not seem to result in better outcomes than conventional rehabilitation. Electrophysiological results suggest that training using a bilateral upper limb robot increases the recruitment of motor neurons rather than improving the conduction function of the corticospinal tract.