Wearable assistive robots need to continuously estimate terrain, locomotion mode, gait phase, and walking speed to adapt their support. According to a new arXiv preprint, these quantities are usually handled by separate models that are each tuned to a specific terrain, and datasets for such systems are rarely collected under realistic conditions.
The paper introduces Tread, a visuo-inertial framework designed to estimate all of these quantities together. By combining visual and inertial data, Tread is intended to be context-aware, meaning its estimates can reflect both the environment and the user's current mode of locomotion.
This unified approach stands in contrast to the common practice of using terrain-specific models. The abstract does not provide details on the architecture or experimental results, so those aspects are not covered here.