Artificial intelligence (AI) tools are increasingly able to learn from larger and more complex data, thus allowing clinicians and scientists to gain new insights from the information they collect about their patients every day. In rehabilitation medicine, AI can be used to find patterns in huge amounts of healthcare data. These patterns can then be leveraged at the individual level, to design personalized care strategies and interventions to optimize each patient’s outcomes. However, building effective AI tools requires many careful considerations about how we collect and handle data, how we train the models, and how we interpret results. In this perspective, we discuss some of the current opportunities and challenges for AI in rehabilitation. We first review recent trends in AI for the screening, diagnosis, treatment, and continuous monitoring of disease or injury, with a special focus on the different types of healthcare data used for these applications. We then examine potential barriers to designing and integrating AI into the clinical workflow, and we propose an end-to-end framework to address these barriers and guide the development of effective AI for rehabilitation. Finally, we present ideas for future work to pave the way for AI implementation in real-world rehabilitation practices.

Repetitive lifting of heavy loads increases the risk of back pain and even lumbar vertebral injuries to workers. Active exoskeletons can help workers lift loads by providing power assistance, and therefore reduce the moment and force applied on L5/S1 joint of human body when performing lifting tasks. However, most existing active exoskeletons for lifting assistance are unable to automatically detect user's lift movement, which limits the wide application of active exoskeletons in factories. In this paper, we propose a simple but effective lift detection strategy for exoskeleton control. This strategy uses only exoskeleton integrated sensors, without any extra sensors to capture human motion intentions. This makes the lift detection system more practical for applications in manufacturing environments. Seven healthy subjects participated in this research. Three different sessions were carried out, two for training and one for testing the algorithm. In the two training sessions, subjects were asked to wear a hip exoskeleton, controlled in transparent mode, and perform repetitive lifting and a locomotion circuit; lifting was executed with different techniques. The collected data were used to train the lift detection model. In the testing session, the exoskeleton was controlled in order to deliver torque to assist the lifting action, based on the lift detection made by the trained algorithm. The across-subject average accuracy of lift detection during online test was 97.97 ± 1.39% with subject-dependent model. Offline, the algorithm was trained with data acquired from all subjects to verify its performance for subject-independent detection, and an accuracy of 97.48 ± 1.53% was achieved. In addition, timeliness of the algorithm was quantitatively evaluated and the time delay was <160 ms across different lifting speeds. Surface electromyography was also measured to assess the efficacy of the exoskeleton in assisting subjects in performing load lifting tasks. These results validate the promise of applying the proposed lift detection strategy for exoskeleton control aiming at lift assistance.

Dynamic control of robotic exoskeletons is paramount to ensuring safe, synergistic assistive action of functional benefit to users. To date, exoskeleton controllers have excelled in rhythmic and quasi-rhythmic tasks, whereas control methods for assisting discrete movements remain limited by their task-specificity. Inspired by neurophysiological dynamic movement primitives (DMPs), we formulated a novel controller that facilitated a variety of lifting movements using a single adaptive DMP (aDMP), for wearable robotic assistance of discrete movements. For a variety of load lifting tasks, we first benchmarked our method's trajectory prediction accuracy against the state-of-the-art DMP using passively recorded exoskeleton sensor data (offline), followed by a functional validation of online aDMP trajectory estimates. Finally, we assessed the functional effects of aDMP-based exoskeletal assistance on joint kinematics and muscular activity during repetitive lifting. The new aDMP method accurately predicted and smoothly synchronized robotic assistance with variable movement trajectories, resulting in reduced muscular activation of the erector spinae muscles (up to 47.6%) while preserving lower-limb joint kinematics and reducing the extension time by 15.5% compared to unassisted conditions. This method holds promise for use in a wide range of wearable robotic applications, including both clinical rehabilitation and user assistance in activities of daily living and/or manual labor.

The number of exoskeletons providing load-lifting assistance has significantly increased over the last decade. In this field, to take full advantage of active exoskeletons and provide appropriate assistance to users, it is essential to develop control systems that are able to reliably recognize and classify the users’ movement when performing various lifting tasks. To this end, the movement-decoding algorithm should work robustly with different users and recognize different lifting techniques. Currently, there are no studies presenting methods to classify different lifting techniques in real time for applications with lumbar exoskeletons. We designed a real-time two-step algorithm for a portable hip exoskeleton that can detect the onset of the lifting movement and classify the technique used to accomplish the lift, using only the exoskeleton-embedded sensors. To evaluate the performance of the proposed algorithm, 15 healthy male subjects participated in two experimental sessions in which they were asked to perform lifting tasks using four different techniques (namely, squat lifting, stoop lifting, left-asymmetric lifting, and right-asymmetric lifting) while wearing an active hip exoskeleton. Five classes (the four lifting techniques plus the class “no lift”) were defined for the classification model, which is based on a set of rules (first step) and a pattern recognition algorithm (second step). Leave-one-subject-out cross-validation showed a recognition accuracy of 99.34 ± 0.85%, and the onset of the lift movement was detected within the first 121 to 166 ms of movement.

Passive ankle-foot prostheses are light-weighted and reliable, but they cannot generate net positive power, which is essential in restoring the natural gait pattern of amputees. Recent robotic prostheses addressed the problem by actively controlling the storage and release of energy generated during the stance phase through the mechanical deformation of elastic elements housed in the device. This study proposes an innovative low-power active prosthetic module that fits on off-the-shelf passive ankle-foot energy-storage-and-release (ESAR) prostheses. The module is placed parallel to the ESAR foot, actively augmenting the energy stored in the foot and controlling the energy return for an enhanced push-off. The parallel elastic actuation takes advantage of the amputee’s natural loading action on the foot’s elastic structure, retaining its deformation. The actuation unit is designed to additionally deform the foot and command the return of the total stored energy. The control strategy of the prosthesis adapts to changes in the user’s cadence and loading conditions to return the energy at a desired stride phase. An early verification on two transtibial amputees during treadmill walking showed that the proposed mechanism could increase the subjects’ dorsiflexion peak of 15.2% and 41.6% for subjects 1 and 2, respectively, and the cadence of about 2%. Moreover, an increase of 26% and 45% was observed in the energy return for subjects 1 and 2, respectively.

Low-back wearable robots are emerging tools to provide support to operators during handling of goods and repetitive operations. In this paper, we present and validate a novel control strategy for an active pelvis orthosis, which operates intuitively and effectively to assist workers during lifting operations. The proposed control strategy has a hierarchical architecture: the first layer (the intention-detection module) aims to detect online the onset of the lifting movement; the second layer (the assistive strategy) computes the reference torque profile to assist the movement, after the movement onset is detected; the third layer (the low-level control layer) aims at setting the current to drive the actuators. The control strategy relies on the angle measurements acquired by the encoders integrated in the robotic device and does not need additional sensors to detect the event. The system was tested on a pool of five healthy subjects, who were requested to perform repetitive lifting movements: first, the subject was requested to bend the trunk, grasp the box, lift it up and place it on a table; second, the subject was requested to grasp the object from the table, lower it down, place it on the floor and get up without the load. The tasks were executed with the exoskeleton controlled in transparent and assistive modes. Results show that the assistive action allows to perform the lifting movement faster. Significant reductions of the activity of the Lumbar Erector Spinae muscles were observed in the assistive mode compared to the transparent mode: a 16% reduction was observed when extending the trunk while holding the weight and a 33% reduction when extending the trunk without holding the load.