The original HARIA project aims to build human sensorimotor augmentation systems, particularly for people with uni- or bi-lateral upper-limb chronic motor disabilities. This new approach to physical human-robot interaction is achieved through the development of supernumerary robotic limbs and wearable sensorimotor interfaces. The combination of these technologies, together with Artificial Intelligence methodologies, allows the users to control and feel the robotic limbs as an extension of their bodies. In this Hop-On Facility activity, we propose to expand the ambition of the HARIA project, by adding a robust layer that manages the occurrence of failures and anomalous situations, using a combination of methods to Predict, Detect, Communicate, and Recover from failure events during the execution of collaborative and augmented tasks, aimed at the application scenarios envisaged in HARIA. Prediction of failure events will be done through a Deep Learning classifier using multi-modal sensory information, while actual failure detection relies mostly on model-based methods. Failures are communicated to the users via the Wearable Sensorimotor Interfaces developed in the project, who can then control the reaction strategies. HARIA-FT strives to improve the robustness of the original system through this failure management framework and through extensive experimentation and validation.