Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Device Technologies and Biomedical Robotics

Device Technologies and Biomedical Robotics - Poster Session C

Poster I14 - Instantaneously Adaptive Exoskeleton Control using Predicted Biological Knee Moment: Real-Time Implementation and Validation

Friday, October 25, 2024
10:00 AM – 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  Many children with cerebral palsy (CP) exhibit crouch gait, a gait deviation where individuals walk with persistently flexed limbs, potentially leading to decreased or limited mobility. Acute observational studies have shown motorized robotic exoskeletons that provide assistive torque during the stance and late swing phases of walking improve mobility and gait biomechanics by increasing knee extension. Our prior work introduced an adaptive controller that adjusts timing and magnitude of knee extension assistance based on estimates of internal knee moment as an alternative to controllers that apply a pre-tuned level of torque within each phase of the gait cycle (Chen et al. 2021). This adaptive controller targets individual effort and allows robotic assistance to be adjusted based on knee moment estimations rather than simply joint position. Modeling the knee as a torsion spring allows accurate and instantaneous knee moment estimations using only changes in knee angle and individual-specific stiffness values.

Previous implementations have demonstrated the feasibility of this approach to accurately estimate internal knee moment and adjust exoskeleton torque in real-time. The objective of this work was to 1) Design and implement real-time software using a flexible Controller Area Network (CAN) communication system to instantaneously adapt applied exoskeleton knee torque proportional to the estimated knee moment during walking and 2) Validate performance in bench-top testing that simulated overground walking. This design validation is a necessary step to create a robust exoskeleton control system that reliably delivers adaptive torque for overground gait assistance and training.

Materials and

Methods:  This study utilized an untethered adjustable exoskeleton with a single actuated degree of freedom at the knee and a passive ankle joint (Figure 1A). The actuator (Agilik, Bionic Power, Vancouver, Canada) can apply bidirectional torque to assist knee flexion and extension. Knee angle and angular velocity were measured by a built-in encoder. A force sensitive resistor (FSR) embedded in the foot plate measured ground contact force. Data acquisition, control, and communication were implemented in a Raspberry Pi Model 4B with a CAN attachment (Figure 1B).

To estimate the user’s biological knee extensor moment the knee was modeled as a dynamic spring by splitting stance into three phases (Figure 2): early-stance (knee flexion after initial contact), mid-stance (knee extension), and late-stance (knee flexion before terminal contact) with a unique stiffness value for each phase (kF_es , kE_ms, and kF_ls, respectively). These stiffness values, as well as knee extensor moment offsets at each transition between states (KEMes, KEMms, KEMls) are individual-specific; here they were calculated using data from healthy individuals (Winters, 2009). Stance phase transitions were determined by angular velocity zero crossings (i.e., switches between flexion/extension) whereas transitions between stance and swing were governed by FSR thresholds (Figure 3). A 0.1-sec deadband was implemented after each transition to prevent incorrect jumps between states.

Ten simulated trials of two strides each were completed in a benchtop setup to evaluate the system design, validate the instantaneous estimate of knee moment from measured knee angle, and quantify the time delay of torque adaptation.

Results, Conclusions, and Discussions: All four gait phases (Figure 3) were clearly identified during 100% of strides taken during benchtop validation testing (Figure 4) which is crucial for accurate knee moment estimation given the piecewise model used. The estimated knee moment profile during stance phases reflected the change in function at the knee; that is, torque alternated between extension and flexion as expected by gait phase. Exoskeleton torque output was scaled to 50% of the estimated moment and adjusted in real-time. Across the 20 strides, the update delay, which quantifies the maximum time delay of the predictive model, was 3.0±0.25 ms (mean±standard deviation) which was an order of magnitude lower than in the prior implementation of the same model in real-time adaptive control (26±0.2 ms, Chen et al. 2021). Because this delay is substantially below the threshold of what is perceivable for a human, these results validate the design goal of providing instantaneously adaptive control. Thus, this study provides impetus for further development and eventual clinical implementation of the designed Raspberry Pi and CANBus interface for exoskeleton control.

There are several limitations in the current study. First, the exoskeleton slew rate was 80 Nm/sec, which limits the maximum rate of change in torque magnitude, but did not affect the system update rate or the predicted knee moment. Another limitation is the need for a kinematic analysis to personalize stiffness and peak knee moment offset values, which considerably affected the accuracy of the knee moment estimations from the piecewise linear model. Here, readily available kinematic data were used rather than personalized data, resulting in discontinuities in knee moment estimations. This did not impact the technical validation of the control architecture, only the accuracy of the predicted moment. Future work should integrate personalized kinematics and compare knee moment estimations to inverse dynamics from ground reaction forces. Improvements in portability and expanding trials to overground walking for healthy individuals and those with crouch gait are warranted. Long-term, the adaptability of this system may enhance efficacy of exoskeletons in real-world scenarios for users with varying gait patterns and needs.

Acknowledgements (Optional):  This work was supported by the intramural research program of the NIH Clinical Center and by the NIH Biomedical Engineering Summer Internship Program (BESIP).