Posters

Poster 1 - The User Command Interface: A Flexible and Adaptable Setup System for Industrial Active Exoskeletons

Olmo A. Moreno F., Daegeun Park, Christian Di Natali, Darwin G. Caldwell, and Jesus Ortiz | Istituto Italiano di Tecnologia

Topics: Human Utilization, User-driven Settings

This study evaluates the flexibility of the User Command Interface (UCI), an adaptable setup system for industrial exoskeletons. This interface allows users to adjust assistive force parameters easily, it offers security-focused operation with minimal distractions and mechanical navigation for harsh environments. It provides access to system configuration parameters, signal visualization, and tutorials. The UCI enhances user adaptability and control in industrial settings, simplifying tasks and facilitating efficient communication with the exoskeleton system. Experiments were conducted with participants using XoTrunk (back support exoskeleton) to modify the force assistance and perform standard lifting activities, and with the upper-limbs exoskeleton Shoulder-SideWINDER. In the last one, the UCI assists the user to proceed with the exoskeleton calibration by running a tutorial and playing audio notifications.

Poster 2 - Simulation and Learning-based Techniques for Exoskeleton Design

Vighnesh Vatsal

TCS Research

*Topics: Actuation & Hardware, Human Utilization *

Exoskeletons are witnessing increasing adoption across industrial and healthcare-related applications. However, their design process still largely follows the conventional approach of iterative physical prototyping along with in-lab usability studies. Recent advances in physics simulators such as MuJoCo have enabled the creation of fast, realistic models of human musculoskeletal systems. We are exploring if these models can provide an alternative to the conventional design process. Instead of in-lab usability studies that employ limited human movements, we directly analyze the exoskeleton’s assistance on a musculoskeletal level in simulation with motion capture data from real tasks. Parameterizing the design of the device and formulating the assistance provided by it into a cost function, we have explored conventional optimization techniques as well as online optimization using reinforcement learning to improve upon existing designs. Validation of this method on real-world designs is an ongoing challenge that would require a multidisciplinary approach to solve.

Poster 3 - Fully FDM-Printed Soft Hand Exoskeleton

Lehong Wang, Zilin Dai, and Markus P. Nemitz

Worcester Polytechnic Institute

Keywords: Actuation & Hardware

Stroke survivors frequently encounter challenges in executing daily living activities due to hand disabilities, necessitating specialized hand assistive technologies, with exoskeletons being the most prevalent. To create customized soft exoskeletons, most designs rely on injection molding techniques, which require lengthy manufacturing and human assembly processes. Other fully 3D-printable exoskeletons are composed only of rigid materials, lacking the ability to conform well to human ergonomics. This work introduces a process to design and fabricate soft pneumatic assistive exoskeletons using commercially available desktop FDM 3D printers. The process consists of a parameterized exoskeleton design that fits the human hand anatomy during actuation and a closed-loop vision-based system to ensure air-tight printing of soft materials. We evaluate our result by conducting experiments on objects requiring different types of grasping motion and demonstrating that the subject is able to hold the objects without exerting voluntary force with their fingers.

Poster 4 - Lower Limb Exoskeleton: Application and Analysis of End-to-End Reinforcement Learning-Based Control

Minsu Kim and Jaeheung Park

Seoul National University & Advanced Institutes of Convergence Techonology (AICT)

Keywords: Control & Modeling

Exoskeleton technology has been evolved for a century, yet challenges in functionality persist. This study introduces an end-to-end Reinforcement Learning (RL) based control method for a hip-assist exoskeleton, aiming to facilitate aperiodic gait assistance. Aperiodic gait is defined as a sequence of stopping (s0) – flat walking (s1) – another stop (s0) – climbing stairs (s2) – final stop (s0). Motion data, acquired from subjects outfitted with an exoskeleton, is learned by the RL-based control network to directly generate control torque. Deep Deterministic Policy Gradient (DDPG) is used as a control network to learn the policy for tracking human motion trajectories and providing assistance as needed, primarily through simulation. The performance of the proposed method is validated both in simulation and in real-world experiments, with the exoskeleton securely mounted on the testbed. Results indicate tracking errors of 0.0548 radians in simulation and 0.0681 radians in real-world settings, as measured by root mean square errors.

Poster 5 - The Delta-Lognormal Law Explains Trunk Velocity Profiles During Lifting and Lowering Trajectories

Ashwin Narayan, Seyram Ofori, Cindy Sia, Yiang Yu, Shounak Bhattacharya, Haoyong Yu

National University of Singapore

Topics: Actuation & Hardware, Sensing, Control & Modeling

Back support exoskeletons have been proposed as a potential solution for critical ergonomic challenges that cause back injuries in many industrial workplaces. However, real-time intent detection and intuitive control of these wearable robotic devices are still a key challenge that impedes adoption, and limits usefulness to those who would benefit from these devices. We discovered that sagittal plane trunk angular velocities during lifting and lowering is explained by the delta-lognormal law. By using this law, we were able to detect start and stop intent for lifting and lowering motions in real-time use it within an assistance strategy for an active back support exoskeleton. Experimental results show (1) close agreement between model and actual trunk velocities, (2) that optimal timing of assistance onset can be measured against the angular velocity and acceleration during lifting. This approach could lead to better, more comfortable exoskeleton controls.

Poster 6 - A Framework for Safe and Natural Mobility in Lower Limb Exoskeletons

Nimrod Curtis, Anis Shakour, David Hexner, Yehuda Bitton and Avishai Sintov

Topics: Sensing, Controls

Robotic exoskeletons have considerable potential to restore mobility in individuals with various disabilities. The technology has not yet been adopted significantly due to several barriers, often related to the complexity of usage and the quality of Human-Robot Interaction. We propose a framework, adopted over the ReWalk exoskeleton by Lifeward Ltd., to cope with the challenges of human-exoskeleton interaction. We aim to minimize the user’s cognitive effort through intuitive interface, intention recognition and context awareness. A stereo camera is employed to detect and model specific obstacles such as stairs, providing crucial geometric parameters including distance, approach angle and height, for safe and seamless traverse in varied environments. Additionally, we introduce an intent recognition module utilizing inertial measurements from the crutch’s IMU to detect transitions, dynamically controlling exoskeleton motors based on geometric parameters upon detecting intent and environment. This integrated approach offers a comprehensive solution for optimizing user assistance and natural mobility.

Poster 7 - Isometric Grip Force Control: A Novel Approach for Improving Control of Hand Exoskeletons

Quentin A. Sanders

George Mason University

Topics: Actuation & Hardware, Controls

Functional limitations imposed on the hands due to impairment from neurological injury may have a substantive impact on overall quality of life of the people impacted. Hand exoskeletons offer promise in aiding individuals with impairments in daily activities, yet their adoption remains low due to challenges in detecting user intent. Recent advancements have brought forth a variety of control methods, with surface electromyography and manual triggers emerging as the predominant strategies, each possessing distinct strengths and inherent weaknesses. We propose the use of an isometric grip force control in which force measured from the ring, and little finger (or middle, ring, and little finger) are used to control an exoskeleton that actuates the remaining digits Initial experiments with able-bodied individuals in two independent studies demonstrate effective utilization of the control strategy to improve task performance, highlighting its ease of use and effectiveness, though further testing with neurological participants is required.

Poster 9 - Forecasting Center of Pressure Location Prior to Foot Strike through Computer Vision and Deep Learning

Michael Murray, James Tung, and Richard Nuckols

University of Waterloo & University of Massachusetts Lowell

Topics: Sensing

Computer-vision has been used for environmental classification during locomotion; however, the ability to predict how the foot will contact a changing environment is underexplored. In this preliminary study, we evaluated the feasibility of forecasting foot center of pressure (COP) prior to foot strike in gait. A simple stair stepping task was performed. A camera (OAK-D) captured images of the foot and stair prior to and at contact with the stair. An instrumented insole (Moticon) recorded the foot COP at contact. Two participants completed the study (n = 111 and 98 steps). A CNN was trained on the temporal camera images with insole COP as ground truth to predict the distal-proximal COP location prior to foot strike. The model prediction at 116ms prior to foot contact had an R2 < 0.74 and an RMSE < 30mm. Future work will involve testing with more participants and broadening the environmental variables.