Summary: A new study details how AI and computer simulations train robotic exoskeletons to help users save energy when walking, running and climbing stairs. This method eliminates the need for lengthy experiments involving humans and can be applied to various assistive devices.
This advancement offers significant potential to help people with mobility issues, by improving accessibility in daily life. The researchers found that participants used up to 24.3% less energy with the exoskeleton.
Highlights:
- AI and simulations train exoskeletons without experiments involving humans.
- The exoskeletons helped users save up to 24.3% energy during movement tests.
- The method can be applied to various assistive devices, including prostheses.
Source: New Jersey Institute of Technology
A team of researchers has demonstrated a new method that leverages AI and computer simulations to train robotic exoskeletons that can help users save energy when walking, running and climbing stairs.
Described in a study published in Naturethe new method rapidly develops exoskeleton controllers to facilitate locomotion without relying on lengthy experiments involving humans.
Additionally, the method can be applied to a wide variety of assistive devices beyond the hip exoskeleton demonstrated in this research.
“This may also apply to knee or ankle exoskeletons, or other multi-joint exoskeletons,” said Xianlian Zhou, associate professor and director of the BioDynamics Lab at NJIT.
In addition, it can also be applied to prostheses above or below the knee, providing immediate benefits to millions of able-bodied and mobility-impaired people, he said.
“Our approach marks a significant advancement in wearable robotics, as our exoskeleton controller is exclusively developed via AI-driven simulations,” explains Zhou. “Additionally, this controller seamlessly transitions to hardware without requiring additional testing on human subjects, making it experiment-free.”
This breakthrough holds promise for helping people with mobility issues, including older adults or stroke survivors, without requiring their presence in a laboratory or clinical environment for extensive testing. Ultimately, it paves the way for restoring mobility and improving accessibility to daily life at home or in the community.
“This work proposes and demonstrates a new method that uses physics and data-based reinforcement learning to control wearable robots to directly benefit humans,” says Hao Su, corresponding author of a paper on the work and associate professor of mechanics. and aerospace engineering at North Carolina State University.
Exoskeletons have the potential to enhance human locomotive performance in a wide variety of users, from injury rehabilitation to lifelong assistance to people with disabilities. However, the length of human testing and control laws have limited its widespread adoption.
Researchers have focused on improving the autonomous control of embedded AI systems, that is, systems in which an AI program is integrated with physical technology.
This work focused on teaching robotic exoskeletons how to assist able-bodied people with a variety of movements, and expands on previous research based on reinforcement learning for lower extremity rehabilitation exoskeletons, also a collaborative effort between Zhou, Su and several others.
“Previous achievements in reinforcement learning have tended to focus mainly on simulation and board games. Our method provides a basis for turnkey solutions in the development of controllers for wearable robots,” explains Shuzhen Luo, assistant professor at Embry-Riddle Aeronautical University and first author of both work. Luo previously worked as a postdoctoral fellow in Zhou’s and Su’s labs.
Normally, users must spend hours “training” an exoskeleton so that the technology knows how much force is needed – and when to apply that force – to help users walk, run or climb stairs.
The new method allows users to use the exoskeletons immediately because the closed-loop simulation integrates both an exoskeleton controller and physical models of musculoskeletal dynamics, human-robot interaction and muscle reactions, thus generating efficient and realistic data and iteratively learning a better control policy in simulation. .
The unit is pre-programmed to be ready for immediate use, and it is also possible to update the controller on hardware if researchers make improvements in the laboratory through extensive simulations. Future prospects of this project include the development of individualized and tailor-made controllers that assist users in various daily life activities.
“This work essentially makes science fiction real, allowing people to expend less energy while performing various tasks,” Su says.
For example, when testing human subjects, researchers found that study participants used 24.3% less metabolic energy when walking with the robotic exoskeleton, compared to walking without an exoskeleton. . Participants used 13.1% less energy when running in the exoskeleton and 15.4% less energy when climbing the stairs.
While this study focused on the researchers’ work with able-bodied people, the new method aims to help people with reduced mobility using assistive devices.
“Our framework can offer a generalizable and scalable strategy for the rapid development and widespread adoption of a variety of assistive robots for able-bodied and mobility-impaired people,” says Su.
“We are in the early stages of testing the performance of the new method on robotic exoskeletons used by elderly people and people with neurological diseases, such as cerebral palsy. And we also want to explore how the method could be used to improve the performance of robotic prosthetic devices.
Funding: This research was carried out with support from the National Science Foundation under grants 1944655 and 2026622; the National Research Institute on Disability, Independent Living and Rehabilitation, under grant DRRP 90DPGE0019; the Swiss Research Fellowship Program of the Administration for Community Living; and the National Institutes of Health, under award 1R01EB035404.
About this research news in AI and neurotechnologies
Author: Déric Raymond
Source: New Jersey Institute of Technology
Contact: Deric Raymond – New Jersey Institute of Technology
Picture: Image is credited to the New Jersey Institute of Technology
Original research: Closed access.
“Experiment-free exoskeleton assistance via simulation learning” by Xianlian Zhou et al. Nature
Abstract
Exoskeleton support without experimentation via simulation learning
Exoskeletons have enormous potential to improve the performance of human locomotives. However, their development and wide distribution are limited by the need for lengthy human tests and artisanal control laws. Here we show a no-experiment method for learning a general-purpose control policy in simulation.
Our simulation learning framework leverages dynamically aware musculoskeletal and exoskeleton models and data-driven reinforcement learning to bridge the gap between simulation and reality without human experiences.
The learned controller is deployed on a custom hip exoskeleton that automatically generates assistance in different activities with reduced metabolic rates of 24.3%, 13.1% and 15.4% for walking, running and stair climbing , respectively.
Our framework can offer a generalizable and scalable strategy for the rapid development and widespread adoption of a variety of assistive robots for able-bodied and mobility-impaired individuals.