This paper presents a novel approach for active gait trajectory planning in lower limb rehabilitation exoskeletons, specifically targeting hemiplegic patients. The proposed method integrates dynamic motion primitives (DMPs) and generalized regression neural networks (GRNN) to accurately simulate lower limb joint motion trajectories. Experimental results demonstrate the superiority of the GRNN-DMPs method over the standalone DMPs approach, as it generates joint trajectories with reduced error and enhanced precision. This approach holds great promise for active rehabilitation training with exoskeleton robots for lower limb rehabilitation.