Gait trajectory planning is a critical component in the control of powered lower limb prostheses. In order to achieve coordination between the prosthesis and the intact limb, existing gait trajectory planning methods generally employ data-driven modeling, which directly maps the movement of intact limbs as the reference trajectory of prostheses. However, these methods often suffer from high modeling complexity and poor perturbation resilience. To address this issue, we proposed a novel gait trajectory planning method driven by the delayed feedback reservoir. In this approach, we utilized the Mackey-Glass oscillator as the nonlinear node of the reservoir, with the hip angle of the amputatied side serving as the input. The output is the mapped knee angle of the prosthesis. Notably, the output of the reservoir is the linear superposition of the virtual node states, offering significant advantages in terms of high global convergence and fast convergence speed during training and computing. Furthermore, we deployed the delayed feedback reservoir on the FPGA hardware and utilized serial communication to achieve data interaction with the STM32 microcontroller, allowing for real-time wearability experiments on a powered lower limb prosthesis. The experimental results show that our model achieves a correlation coefficient of 0.8377 between intact limb and prosthesis under normal walking and 0.7436 under perturbation, demonstrating a strong correlation. The jerk value also reflects the model’s robust resistance to perturbations, with an average jerk of 47 979 deg/s3, which is approximately 31% lower than that of the intact limb. This demonstrates that DFR possesses significant perturbation resistance, offering a new solution for prosthesis control.