Guiding the visually impaired in complex environments requires real-time two-way interaction and safety assurance. We propose a Force-Compliance Model Predictive Control (FC-MPC) and Robot-User Control Barrier Functions (CBFs) for force-compliant navigation and obstacle avoidance in Hexapod guide robots. FC-MPC enables two-way interaction by estimating user-applied forces and moments using the robot’s dynamic model and the recursive least squares (RLS) method, and then adjusting the robot’s movements accordingly, while Robot-User CBFs ensure the safety of both the user and the robot by handling static and dynamic obstacles, and employ weighted slack variables to overcome feasibility issues in complex dynamic environments. We also adopt an Eight-Way Connected DBSCAN method for obstacle clustering, reducing computational complexity from O($n^{2}$ ) to approximately O(n), enabling real-time local perception on resource-limited on-board robot computers. Obstacles are modeled using Minimum Bounding Ellipses (MBEs), and their trajectories are predicted through Kalman filtering. Implemented on the HexGuide robot, the system seamlessly integrates force compliance, autonomous navigation, and obstacle avoidance. Experimental results demonstrate the system’s ability to adapt to user force commands while guaranteeing user and robot safety simultaneously during navigation in complex environments. Note to Practitioners—Guiding visually impaired individuals in complex environments remains a significant challenge, as traditional methods like guide dogs are limited by high costs and scalability issues. To address this, we propose a Hexapod guide robot framework that integrates Force-Compliance Model Predictive Control (FC-MPC) and Robot-User Control Barrier Functions (CBFs) to ensure real-time two-way interaction and dual safety for both the user and the robot. The framework supports diverse interaction modes by allowing parameter adjustments to combine functionalities such as force compliance, global autonomous navigation, and obstacle avoidance. An efficient clustering algorithm is adopted in the local environment perception process, which enables real-time deployment on resource-constrained devices of guide robot. While effective in experiments, future work will focus on map-less navigation to improve performance in previous unseen environments.