<p>The efficacy of upper limb exoskeleton-assisted motion training is critical for the rehabilitation of motor impairments resulting from neural injuries. Despite technological advancements, these devices continue to face substantial limitations in joint safety design, particularly in achieving optimal alignment between the human shoulder joint and the exoskeleton, which is essential for effective and safe rehabilitation. To address these challenges, this study introduces a novel hybrid shoulder rehabilitation exoskeleton that incorporates a follower mechanism and an adaptive regulation algorithm for shoulder redundant force perception. Theoretical analyses reveal that, compared to conventional exoskeleton structures, this mechanism significantly enhances human-robot interaction (HRI) compatibility by improving joint alignment and reducing mechanical constraints. Experimental results demonstrate that the activation of the follower mechanism reduces redundant force at the shoulder joint to 50.1% of those observed in its absence. These findings provide robust evidence that the proposed design not only mitigates shoulder redundant forces but also supports a wide range of rehabilitation training modalities.</p>