Abstract: Existing walker navigation algorithms mainly focus on passive obstacle avoidance, which makes it difficult to satisfy the demands of active rehabilitation training while ensuring safety. To address this contradiction, this paper proposes a rehabilitation-guided potential field method based on dynamic human–robot weighting. By interpreting the user’s desired velocity and current motion input, the proposed method adaptively adjusts the intention weight, thereby balancing user autonomy and machine assistance within a shared-control framework. Unlike conventional methods that use potential fields solely for obstacle avoidance, the proposed method innovatively redefines the potential field as a guidance tool for rehabilitation training: the attractive potential field is dynamically modulated based on the patient's capability, while the repulsive potential field enables patient exploration within safe boundaries. Finally, under the action of the smoothing and prediction terms in the designed model predictive controller (MPC), Through its receding-horizon optimization, this control framework is capable of producing personalized reference trajectories that are both safe and easy to track., thereby balancing immediate safety and long-term rehabilitation training.MATLAB simulation results show that the proposed method achieves a success rate of up to 97.4% and significantly improves path smoothness during turning and obstacle avoidance. Real-world scenario tests further validate the feasibility and effectiveness of this control strategy in typical rehabilitation settings, demonstrating its potential for clinical application.