Our research focuses on enabling symbiotic relationship between human wearers and wearable robotics. We envision a computation foundation and engineering framework that enables seamless integration and co-adaptation of smart wearable machines (such as robotic prostheses, exoskeletons, functional electrical stimulation) and humans to assist and augment human physical abilities, performance, and experiences.
We plan to achieve our vision via study of wearer-machine interaction and development of breakthrough technologies in wearable robots and neural-machine interfaces. The wearer-machine integrated system needs to achieve the following functions.
- Environment Adaptation
- Personalization of Wearable Robots and Wearer-Robot Co-Adaptation
Recently, we discussed our research and perspectives on AI applications in robotic prostheses on Science. Here is the article “Artificial intelligence meets medical robotics“.
Neural Control of Prosthetic Legs (Read our recent perspective paper)
The function of current robotic prosthetic legs is limited due to the lack of neural-machine interface (NMI). NREL pioneered a novel phase-dependent, neuromuscular-mechanical fusion-based NMI for powered artificial legs. The NMI recognizes the user’s locomotion mode and predict the locomotion mode transitions. This basic engineering framework has been published in 2009 and 2011; it has been widely adopted since then. The technology has enabled individuals with lower limb amputations to walk with robotic prosthetic devices intuitively and negotiate changing terrains smoothly and seamlessly.
Our recent research effort focuses on restoring anticipatory and compensatory postural control in individuals with transtibial amputations via feed-forward, continuous neural control of robotic prosthesis ankle. Our recent paper published in Science Robotics has shown that this approach transforms the way amputees interact with their powered prostheses and restore the near-normal neuromechanics of amputees in postural control.
Related news and videos: Check out our discussion about this research topic at
Check out the video of our study published in Science Robotics at
Neural Control of Prosthetic Arms
Neural control of prosthetic arms is essential for prosthesis embodiment. Our major contributions along this research line include (1) enhancing the robustness and practical value of EMG pattern recognition-based prosthesis control (machine learning-based approach), (2) developing EMG-driven musculoskeletal model for prosthetic arm control (physiological model-based approach), and (3) our current innovation on combination of machine learning and known physical model for myoelectric prosthesis arm control. Our methods have been applied to patients with below elbow amputations, as well as amputees with targeted muscle reinnervation (TMR) surgeries. Currently, we are developing different smart control strategies for dexterous prosthesis hands and wrists in order to reduce the user’s cognitive load in operating the prosthesis in activities of daily living.
Related news and videos: Check out our research news at NC State and
Check out our recent idea that combines AI and physical dynamic model
Reinforcement Learning Control of Prosthetic Legs for Personalized Walking Assistance
Personalization of assistive wearable machines, such as exoskeletons and robotic prostheses, is usually done in clinics manually and heuristically. Our lab has developed several breakthrough technologies that can automatically tuning 12 control parameters simultaneously for robotic prosthetic legs in order to provide personalized walking assistance. Our first approach is to design a cyber expert system that learns to tune a robotic prosthesis from certified prosthetists. The research results were published in 2013 and 2016. Th related US patent has been issued recently. Our second approach is based on reinforcement learning-based optimal adaptive control. This approach does not depend on any prior knowledge on prosthesis tuning. The detailed design can be found in our publications (2017, 2019a, 2019b, 2021 TRO). Both approaches resulted in learned tuning policy that may be directly deployed in the clinics for automatic robotic prosthesis tuning. Our innovation may transform the current clinical practice and improve the function and quality of life for individuals with lower limb amputations.
Our recent study focuses more on how to integrate human gait performance, such as gait symmetry (2022)
Related news and videos: Check out our research news at IEEE Spectrum and
Reinforcement Learning Control of Hip Exoskeleton for Personalized Walking Assistance
Building upon our innovation and success of RL-based prosthesis personalization studies, we further extended this technology to robotic exoskeletons. RL is a powerful, time/data efficient human-in-the-loop learning and optimization method to enable personalized exoskeleton assistance in walking. Please check out our recent studies and publications on this topic (Tu 2021, Nalam 2022, Zhang 2022). We are also working on applying this technology into clinical rehabilitation applications.
Wearer-Robot Interaction, Coordination, and Co-Adaptation (Read our perspective and opinion paper)
Understanding how human interacts with rehabilitation machine is critical to develop effective assistive or therapeutic training tools for medical rehabilitation. Our contribution in this area include (1) investigation of the effects of lower limb prosthesis control errors on the amputees’ walking stability (2015a, 2015b) and perceived stability (2023) and how human wearers’ reaction mitigate these error effects (2022), (2) Understanding the influence of prosthesis mechanics on wearer’s gait performance (2020a, 2020b), (3) studying how augmented feedback and virtual reality can enhance the task performance of individuals with lower limb amputations (2019a, 2019b), and (4) study of human attention, intention, and cognitive work load when interacting with different robotic prostheses in various environments and task contexts.
Our recent innovation focuses on wearer-robot coordination and co-adaptation. We used RL to make robot adaptive to the user; and used visual feedback to enable human adaptation while walking with the robotic prosthesis. The idea has been explored initially (2021). We hope the research can lead to a new paradigm for wearer-robot symbiosis.
Related news and videos: Check out our research news at NC State and
We designed a new hip exoskeleton. Our innovation lies in the design of active hip abduction/adduction joints, beyond the powered hip flexion/extension, in order to assist gait and dynamic postural stability at the same time. In human walking, control of foot placement in both the mediolateral and anteroposterior directions has long been recognized as an effective mechanism for maintaining gait stability. During walking, beyond the forward step length regulated by hip flexion/extension, adaptation of the step width (2024 Featured Paper in TBME, 2023IROS) can be adjusted by hip abduction/adduction assistance. More research will be conducted on individuals with neuromuscular deficits to evaluate the hip exoskeleton design.
Related news and videos: Check out our IEEE TBME featured paper that using a hip exoskeleton to modulate stepwidth