FSGP: Fourier Series based Gait Primitives for Individualized Gait Prediction with Varying Walking Speeds

Walking assistance with normal gait patterns can be provided by Lower Limb Exoskeletons (LLEs) for patients with gait impairments. Since the gait patterns are highly correlative to the anthropometric features of the subject and the walking speed, it is challenging for LLEs to generate individualized gait patterns for different subjects with varying walking speeds. In this paper, the Fourier Series based Gait Primitives (FSGP) and Multi-Layer Perceptron (MLP) neural networks based individualized gait patterns prediction approach is proposed for different subjects in varying walking speed cases. A gait database with the joint angles collected from 30 healthy subjects at 12 different walking speeds was established, as well as an anthropometric database with anthropometric features measured from these subjects was established. The parameters for FSGP and MLP models can be learned off-line based on the gait database and the anthropometric database, and then the individualized joint angles can be predicted online with the specified subject. Experimental results indicate that the reconstructed joint angles are very similar to the subject's actual joint angles for varying walking speeds from 1.0 km/h to 5.0 km/h. Compared with the other methods in the previous studies, the proposed approach can predict individualized gait patterns with less prediction errors.