In the paper, the development, control, and preliminary evaluation of the human–robot coupled walking exoskeleton for weight-support enhancement are presented, which provides the assistance of abduction/adduction and flexion/extension for the hip joint, and the flexion/extension for knee joint of human legs during walking. The trajectory generation strategy for the walking exoskeleton utilizes the inverted pendulum approximation. Considering human in the control loop, the periodic walking trajectory and uncertainties with known periods, we propose a human-cooperative adaptive fuzzy strategy combing virtual tunnels that allows human subjects to change the movement timing of their legs and produce a physiological path. The strategy does not need human–robot coupled model, and the designed compliant virtual constraints keep the legs of the subject within the constrained tunnel around the expected path. The subjects with the assistance of supporting torques can walk effortlessly along the spatial path. The path control strategy has been verified with two healthy subjects. The recorded kinematic data demonstrated that the participants are able to achieve stable walking with larger spatio-temporal variability.