<p>Exoskeletons have exhibited increasingly diverse designs and broader applications in rehabilitation and medical fields. To achieve optimal assistance performance, it is essential that the assistive force be synchronized with human biomechanics. However, current soft exoskeletons still face challenges in achieving natural gait synchronization and providing stable, comfortable assistance. This study proposes a biomimetic assistance method for a hip soft exoskeleton that better matches natural human gait. It explores how integrating a dual‑pretension mechanism with biomechanics‑based force modeling can enhance assistive performance, improve user comfort, and reduce metabolic energy consumption during walking. By mimicking the muscle force characteristics of hip joint extension and flexion, two sets of assistive torque curves were developed to correspond with human biomechanical motion. Additionally, to compensate for the hysteresis inherent in the exoskeleton system, a pre‑tension force was applied before and after the assistive curves to improve response time. To enhance the accuracy of gait cycle prediction and achieve better synchronization with natural gait patterns, a Gaussian-weighted moving average algorithm was employed to adaptively assign higher weights to recent gait data, thereby improving the responsiveness and adaptability of the exoskeleton. In the experiments, six subjects participated, and their net metabolic rates were compared under assisted and unassisted conditions. The results showed that the subjects’ average metabolic cost decreased by 16.4% at a walking speed of 3 km/h and by 14.1% on a 4° slope. Compared with previous approaches, the proposed algorithm achieved more accurate gait‑phase adaptation and reduced metabolic expenditure, highlighting its potential for human–exoskeleton co‑adaptation.</p> Graphical abstract <p></p>