Proprioceptive impairment is a common feature of Parkinson's disease (PD). Proprioceptive function is only partially restored with anti-parkinsonian medication or deep brain stimulation. Behavioral exercises focusing on somatosensation have been promoted to overcome this therapeutic gap. However, conclusive evidence on the effectiveness of such somatosensory-focused behavioral training for improving somatosensory function is lacking. Moreover, it is unclear, if such training has any effect on motor performance in PD. To investigate, whether proprioception improves with a somatosensory focused, robot-aided training in people with PD (PWPs), and whether enhanced proprioception translates to improved motor performance. Thirteen PWPs of mild-moderate clinical severity were assessed and trained ON medication using a robotic wrist exoskeleton. Thirteen healthy elderly participants served as controls. Training involved making increasingly accurate, continuous, precise small amplitude wrist flexion/extension movements. Wrist position sense acuity, as a marker of proprioception function, and spatial error during wrist pointing, as a marker of untrained motor performance, were recorded twice before and once after training. Functional hand writing kinematics exhibited during training were evaluated in the PD group for determining training-induced changes. Training improved position sense acuity in all PWPs (mean change: 28%; < 0.001) and healthy controls (mean change: 23%; < 0.01). Second, 10/13 PD participants and 10/13 healthy control participants had reduced spatial movement error in the untrained wrist pointing task after training. Third, spatial error for the functional handwriting tasks (line tracing and tracking) did not improve with training in the PD group. Proprioceptive function in mild to moderate PD is trainable and improves with a somatosensory-focused motor training. Learning showed a local transfer within the trained joint degree-of-freedom as improved spatial accuracy in an unpracticed motor task. No learning gains were observed for the untrained functional handwriting task, indicating that training may be specific to the trained joint degree-of-freedom.