This study proposes a personalized resistance training control method for wrist rehabilitation robots to address the need for individualized isotonic and isokinetic training in late-stage rehabilitation. Using the NPU-WRIST robot, fuzzy rules were constructed based on joint torque and range of motion to generate personalized resistance and speed parameters. Comparative experiments showed that the personalized scheme achieved significantly higher training efficiency (90.1% for isotonic, 95.3% for isokinetic) than the fixed-parameter scheme (30.5% and 24.8% respectively), verifying its effectiveness.