M. Kim*, Y. Ding*, P. Malcolm, J. Speeckaert, C. Siviy, C. Walsh, and S. Kuindersma. 2017. “
Human-in-the-loop Bayesian optimization of wearable device parameters.” PloS one, 12, 9, Pp. e0184054.
Abstract
The increasing capabilities of exoskeletons and powered prosthetics for walking assistance have paved the way for more sophisticated and individualized control strategies. In response to this opportunity, recent work on human-in-the-loop optimization has considered the problem of automatically tuning control parameters based on realtime physiological measurements. However, the common use of metabolic cost as a performance metric creates significant experimental challenges due to its long measurement times and low signalto-noise ratio. We evaluate the use of Bayesian optimizationÐa family of sample-efficient, noise-tolerant, and global optimization methods-for quickly identifying near-optimal control parameters. To manage experimental complexity and provide comparisons against related work, we consider the task of minimizing metabolic cost by optimizing walking step frequencies in unaided human subjects. Compared to an existing approach based on gradient descent, Bayesian optimization identified a near-optimal step frequency with a faster time to convergence (12 minutes, p < 0.01), smaller inter-subject variability in convergence time (±2 minutes, p < 0.01), and lower overall energy expenditure (p < 0.01)
1_kim_2017_sf_human_in_the_loop_optimization.pdf M. Kim and S. H. Collins. 2017. “
Once-per-step control of ankle push-off work improves balance in a three-dimensional simulation of bipedal walking.” IEEE Transactions on Robotics, 33, 2, Pp. 406-418.
Abstract
Individuals with lower-limb amputation have high fall risk, which could be partially due to a lack of stabilizing control in conventional prostheses. Inspired by walking robots, we hypothesized that modulating prosthetic ankle push-off could help improve amputee balance.We developed a three-dimensional walking model, found limit cycles at two speeds, and designed state-feedback controllers that made once-per-step adjustments to ankle push-off work, fore-aft and medial-lateral foot placement, and ankle roll resistance. To assess balance, we applied increasing levels of random changes in ground height and lateral impulses until the model fell down within one hundred steps. Although foot placement is known to be important to balance, we found that push-off control was at least twice as effective at recovering from both disturbances at both speeds. Push-off work affected both fore-aft and mediolateral motions, leading to good controllability, and was particularly well-suited to recovery from steps up or down. Our results suggest that discrete control of ankle push-off may be more important than previously thought, and may guide the design of robotic prostheses that improve balance.
2_kim_2017_tro_simulation_simplified_model.pdf M. Kim and S. H. Collins. 2017. “
Step-to-step ankle inversion/eversion torque modulation can reduce effort associated with balance.” Frontiers in Neurorobotics, 11, Pp. 62.
Publisher's VersionAbstractBelow-knee amputation is associated with higher energy expenditure during walking, partially due to difficulty maintaining balance. We previously found that once-per-step push-off work control can reduce balance-related effort, both in simulation and in experiments with human participants. Simulations also suggested that changing ankle inversion/eversion torque on each step, in response to changes in body state, could assist with balance. In this study, we investigated the effects of ankle inversion/eversion torque modulation on balance-related effort among amputees (N = 5) using a multi-actuated ankle-foot prosthesis emulator. In stabilizing conditions, changes in ankle inversion/eversion torque were applied so as to counteract deviations in side-to-side center-of-mass acceleration at the moment of intact-limb toe off; higher acceleration toward the prosthetic limb resulted in a corrective ankle inversion torque during the ensuing stance phase. Destabilizing controllers had the opposite effect, and a zero gain controller made no changes to the nominal inversion/eversion torque. To separate the balance-related effects of step-to-step control from the potential effects of changes in average mechanics, average ankle inversion/eversion torque and prosthesis work were held constant across conditions. High-gain stabilizing control lowered metabolic cost by 13% compared to the zero gain controller (p = 0.05). We then investigated individual responses to subject-specific stabilizing controllers following an enforced exploration period. Four of five participants experienced reduced metabolic rate compared to the zero gain controller (−15, −14, −11, −6, and +4%) an average reduction of 9% (p = 0.05). Average prosthesis mechanics were unchanged across all conditions, suggesting that improvements in energy economy might have come from changes in step-to-step corrections related to balance. Step-to-step modulation of inversion/eversion torque could be used in new, active ankle-foot prostheses to reduce walking effort associated with maintaining balance.
kim_2017_inversion_torque_on_balance.pdf