PaperlayerView exact revision on arXiv

Source-linked AI summary

Effects of Assistance Delay on Joint Mechanics and Energetics in Biological Torque Control of a Hip Exoskeleton

Jimin An, Ryan Lee, Jingshu Peng, Eni Halilaj, Inseung Kang

arXiv:2609.25417v1cs.RO

TL;DR

The paper examines whether a fixed biological-torque mapping provides effective hip assistance across distinct walking tasks. It evaluates assistance delays during level-ground, ramp-ascent, and ramp-descent walking, finding that delay effects differ between mechanical and metabolic outcomes.

  • Problem

    It remains unclear whether a fixed state-to-torque mapping provides effective assistance across tasks with distinct mechanical demands.

    Source: §I · p. 1

  • Method

    The study tested assistance-delay effects on biomechanical, metabolic, and subjective outcomes during level-ground, ramp-ascent, and ramp-descent walking.

    Source: §I · p. 2

  • Results

    Mechanical unloading generally decreased as delay increased, while metabolic rate remained comparatively insensitive to delay.

    Source: §V · p. 5

  • Takeaways & Limitations

    State-to-torque mappings should be evaluated against the intended outcome and the assisted joint’s mechanical role.

    Source: §VI · p. 7

    With eight participants, sensitivity to task-dependent and subjective effects was limited, so null results do not establish equivalent delay responses or an absence of systematic preference.

    Source: §V · p. 7

Abstract

from arXiv · show

Biological torque control directly maps an estimated human joint moment to exoskeleton assistance, providing a task-agnostic strategy for supporting diverse locomotor activities. However, it remains unclear whether a fixed state-to-torque mapping provides effective assistance across biomechanically distinct tasks. We examined how assistance delay affected hip exoskeleton performance during level-ground (LG), ramp-ascent (RA), and ramp-descent (RD) walking. Eight participants completed a zero-torque baseline condition and five active assistance conditions with delays ranging from 40 to 320 ms. Across tasks and active delays, assistance reduced net metabolic rate by 5.24%, positive biological hip joint work by 5.86%, and total lower-limb positive joint work by 1.68% (all p < 0.05). Assistance delay affected both joint-work outcomes (both p < 0.001) but not net metabolic rate. Mechanical unloading generally decreased with increasing delay, whereas metabolic benefits remained comparatively stable. Relative to the zero-torque condition, net metabolic rate decreased by 9.75% during LG and 7.20% during RA but increased by 1.23% during RD. We did not detect task-dependent differences in the delay response. Our findings indicate that biological torque mappings should be evaluated based on the target outcome and mechanical role of the assisted joint, and that predominantly positive-power assistance may not generalize to negative-work-dominant locomotion without modification.

I. INTRODUCTION

Biological torque control maps estimated joint moments directly to assistance, avoiding explicit locomotor-state identification. Whether its fixed mapping works across biomechanically distinct tasks remains unresolved, motivating an evaluation of assistance delay across three walking tasks.

I. INTRODUCTION

  • Conventional controllers estimate locomotor states before generating assistance, which may degrade during transitions and non-cyclic movements and is impractical to tune across real-world user states.Biological torque control instead maps an estimated joint moment directly to exoskeleton torque without explicitly identifying locomotor state.Source: §I · p. 1
  • Fixed scaling and delay may limit task-generalizable assistance because mapping parameters affect assistance benefits and tasks impose distinct mechanical demands.Ramp ascent requires greater positive hip work than level walking, whereas ramp descent involves substantial negative work.Source: §I · p. 1
  • Prior evidence is insufficient to determine whether biomechanically distinct tasks require different state-to-torque mappings.A pilot found similar favorable delays for level and ramp-ascent walking but had a small sample; a later optimization study found no significant task-specific advantage.Source: §I · p. 1
  • The study tested how assistance delay affects biomechanical, metabolic, and subjective outcomes during level-ground, ramp-ascent, and ramp-descent walking.The authors hypothesized that delay effects would differ by outcome and task, and explored subjective ratings.Source: §I · p. 2

II. ROBOTIC HIP EXOSKELETON

The bilateral hip exoskeleton uses thigh-mounted IMUs and a biological torque controller to deliver assistance. The controller estimates each limb’s hip moment, then scales, delays, and filters it to produce exoskeleton torque.

A. Hardware Design

  • The bilateral exoskeleton provides hip flexion and extension assistance, with hip-mounted actuators, adjustable user interfaces, and thigh-mounted IMUs for kinematic information.It weighs 4.3 kg and is designed to provide up to 18 Nm of hip torque.Source: A. Hardware Design · p. 2

B. Biological Torque Controller

  • The mapping uses a fixed scaling factor of 0.2 in all assisted conditions and caps commanded torque at 18 Nm.The delay parameter varies by condition; filtering is applied to the scaled, delayed estimate.Source: B. Biological Torque Controller · p. 2
  • Separate temporal convolutional networks estimated hip moments for each locomotor task using data from 10 healthy participants.Each model used six channels from a thigh-mounted IMU and a 950 ms input window; the task-specific models were validated on main-experiment data.Source: B. Biological Torque Controller · p. 2
  • The controller estimates biological hip moments independently for each limb from IMU inputs, then maps each estimate to exoskeleton torque.The mapping scales, delays, and filters the estimate.Source: Figure 1 · p. 2, B. Biological Torque Controller · p. 2

A. Experiment Design

Eight healthy adults completed level-ground, ramp-ascent, and ramp-descent treadmill walking at task-specific speeds. The protocol compared no-exoskeleton and zero-torque baselines with five active-delay conditions, while measuring metabolic, biomechanical, and subjective outcomes.

A. Experiment Design

  • Eight healthy adults walked at 1.2 m/s for level-ground walking and 0.8 m/s for ramp ascent and descent on inclines of +10° and −10°, respectively.Speeds were selected to maintain comparable physiological demands and minimize anaerobic metabolism.Source: A. Experiment Design · pp. 2–3
  • Each task included a no-exoskeleton condition, a zero-commanded-torque condition, and five assisted conditions with total delays from 40 to 320 ms in 70 ms increments.The tested range was bounded by 40 ms of system latency and pilot-reported discomfort above 320 ms during level-ground walking.Source: A. Experiment Design · p. 3
  • Participants completed six minutes of walking per exoskeleton condition, with four minutes of seated rest afterward and 20 minutes between tasks.A standing trial measured resting metabolic rate; task order and exoskeleton-condition order were randomized.Source: A. Experiment Design · p. 3
  • Metabolic rate, joint kinematics and kinetics, and post-trial assistance preference were measured during the protocol.Preference ratings considered comfort and assistance benefit and were made without direct comparison among conditions.Source: A. Experiment Design · p. 3

B. Data Processing and Analysis

The analysis derived lower-limb joint mechanics from motion capture and ground-reaction forces, separated biological hip moment from exoskeleton torque, and quantified positive joint work. Outcomes were compared across assistance delays and tasks.

B. Data Processing and Analysis

  • OpenSim processing combined motion-capture trajectories and treadmill ground-reaction forces to calculate lower-limb joint kinematics and kinetics.The analysis used inverse kinematics and inverse dynamics, with joint-angle and joint-moment trajectories filtered at 4 Hz.Source: 1) Biomechanical Outcomes · p. 3
  • Biological hip moment was calculated by subtracting recorded exoskeleton torque from the net inverse-dynamics hip moment, and positive joint work was integrated over each gait cycle.Total lower-limb positive joint work summed positive hip, knee, and ankle work; 100 gait cycles were sampled from each trial’s final two minutes.Source: 1) Biomechanical Outcomes · p. 3
  • Figure 3 displays task-matched percent changes in metabolic and positive joint-work outcomes, with task symbols and delay-response model predictions.Its markers show means and SEM across eight participants; asterisks indicate corrected paired-test differences from NoAssi.Source: Figure 3 · p. 4
  • For one participant’s 250 ms level-ground trial, missing recorded exoskeleton torque was reconstructed from motion-capture-derived hip moments for biomechanical analysis.Source: 1) Biomechanical Outcomes · p. 4

2) Metabolic Rate:

Metabolic rate was calculated from respiratory data collected during the final two minutes of each trial. Net metabolic rate and assistance responses were then derived relative to resting and task-specific baseline values.

2) Metabolic Rate:

  • Breath-by-breath respiratory data from the last 2 min of each trial were used to compute gross metabolic rate normalized by body mass.The standing-trial resting metabolic rate was subtracted to obtain net metabolic rate for subsequent analyses.Source: 2) Metabolic Rate · p. 4
  • Active-condition metabolic and biomechanical values were expressed as percent changes relative to the task-specific NoAssi condition.For each participant and outcome, the overall assistance response averaged these changes across five active delays and three locomotor tasks.Source: 2) Metabolic Rate · p. 4
  • The overall assistance response was calculated by averaging each participant’s percent changes across the five active delays and three locomotor tasks.Source: 2) Metabolic Rate · p. 4

C. Statistical Analysis

The analysis tested whether delay responses varied by outcome and locomotor task, then characterized outcome-specific and exploratory linear trends. Subjective ratings were analyzed separately, and Fig. 4 displays ratings by task and delay.

C. Statistical Analysis

  • A three-way repeated-measures ANOVA tested Outcome, Task, and Delay effects on NoAssi-relative percent changes, with interactions assessing H1 and H2.Outcome × Delay tested whether delay responses differed among outcomes; Task × Delay tested task dependence, and the three-way interaction tested whether task dependence varied by outcome.Source: C. Statistical Analysis · p. 4
  • Separate two-way repeated-measures ANOVAs characterized each outcome, followed by paired t-tests with Bonferroni correction for significant main effects.The pairwise tests used participant-level marginal means.Source: C. Statistical Analysis · p. 4
  • Exploratory linear mixed-effects models treated delay as a continuous predictor scaled per 70 ms and tested overall and task-dependent trends.Models included Delay, Task, and their interaction as fixed effects, Subject as a random intercept, and likelihood-ratio tests of nested models.Source: C. Statistical Analysis · p. 4
  • Objective assistance effects were tested against zero, and each active delay was compared with task-specific NoAssi using paired tests corrected across five delays.Subjective ratings were separately analyzed with a Task-by-Delay repeated-measures ANOVA and corrected pairwise comparisons.Source: C. Statistical Analysis · p. 4
  • The study defined significance as p < 0.05 and estimated post hoc power and sample sizes for 80% power using 20000 simulations.Simulations reflected observed effect patterns and within-subject covariance structure.Source: C. Statistical Analysis · p. 4
  • Fig. 4 arranges ratings by task, connects each participant across delays, and shows group means with ± SEM across eight participants.Ratings used a five-point preference scale combining perceived comfort and assistance benefit; higher scores indicated more favorable assistance.Source: Figure 4 · p. 5

IV. RESULTS

Assistance reduced all three objective outcomes overall, but increasing delay weakened mechanical unloading without a corresponding metabolic trend. Delay-response patterns did not differ detectably by task, though metabolic benefits varied across tasks.

A. Metabolic and Biomechanical Outcomes

  • 5.24% lower net metabolic rate, 5.86% lower positive biological hip joint work, and 1.68% lower total lower-limb positive joint work occurred with assistance versus NoAssi.These averages were across tasks and active-delay conditions; all three reductions were significant.Source: A. Metabolic and Biomechanical Outcomes · p. 5
  • As delay increased, reductions in positive biological hip joint work and total lower-limb positive joint work decreased, while net metabolic rate showed no linear delay-response trend.The mixed-effects trends were significant for both joint-work outcomes (both p < 0.001), but not for net metabolic rate (p = 0.577).Source: A. Metabolic and Biomechanical Outcomes · p. 5
  • No task-dependent delay response was detected; net metabolic rate decreased by 9.75% in LG and 7.20% in RA, but increased by 1.23% in RD.The Task × Delay interaction was not significant in omnibus or outcome-specific analyses.Source: A. Metabolic and Biomechanical Outcomes · p. 5
  • At least one delay reduced each outcome versus NoAssi in LG and RA, but none did so in RD.The largest observed reductions in both positive joint-work outcomes occurred at 40 ms during LG and RA; largest metabolic reductions were 11.33% at 180 ms in LG and 8.97% at 110 ms in RA.Source: A. Metabolic and Biomechanical Outcomes · p. 5

B. Subjective Outcomes

Subjective ratings varied across participants, with no detected group-level task, delay, or interaction effect. The highest mean rating occurred at different delays in each task.

B. Subjective Outcomes

  • The highest mean ratings occurred at 250 ms in LG, 180 ms in RA, and 40 ms in RD, but no Task, Delay, or interaction effect was detected.Ratings varied substantially across participants and did not identify a consistent group-level preference for an assistance delay.Source: B. Subjective Outcomes · p. 5

V. DISCUSSION

Assistance delay affected mechanical unloading more than metabolic benefit, and outcome and task context shaped how the fixed mapping performed. The discussion recommends evaluating mappings against the assisted joint’s mechanics and the intended outcome, while treating null task differences cautiously.

V. DISCUSSION

  • Positive biological hip joint work and total lower-limb positive joint work reductions generally declined as delay increased, while net metabolic rate stayed relatively insensitive.Positive joint work reflects the mechanical response to torque, whereas metabolic rate integrates muscle activation, force production, stabilization, and compensatory responses. Despite a wider tested delay range, this pattern was similar to a prior level-ground study.Source: §V · pp. 5–6
  • The delay minimizing metabolic rate may not maximize positive biological joint-work reductions; select parameters by intended application and consider both outcomes when both matter.Prior optimization studies have used metabolic cost as the primary objective, –.Source: §V · pp. 5–6
  • Across LG and RA, multiple delays yielded benefits, but no delay reliably reduced evaluated outcomes in RD; its energy-absorption demands may mismatch positive-power assistance.The proposed mismatch is a biomechanical explanation, not a directly tested mechanism.Source: §V · p. 6
  • Nonsignificant task-by-delay effects do not establish equivalent task mechanics; Powell et al. likewise found no added task-specific metabolic benefit across evaluated walking conditions.Powell et al. reported metabolic reductions of 8.6% and 8.4% for task-specific and task-agnostic assistance, respectively. The interaction results address delay sensitivity, distinct from the overall effectiveness of the tested mapping.Source: §V · p. 6
  • Task-specific mapping need may track overlap in assisted-joint mechanics more than locomotor labels; because mechanics are joint-specific, hip generalization may not transfer to knee.At the hip, graded walking mechanics are modulated predominantly through positive work, whereas knee mechanics are modulated more strongly through negative work,.Source: §V · p. 6
  • Subjective preference is complementary, not a surrogate; active-delay metabolic differences fell below the average just-noticeable difference of 22.7±17.0% at a 75% discrimination threshold.Ratings may also reflect comfort and synchronization with the device.Source: §V · pp. 6–7
  • Eight participants limited sensitivity to task-dependent and subjective effects; delay-only tuning neither identifies a precise optimum nor tests the proposed RD negative-work mismatch.Post hoc power for categorical task-dependent and subjective effects ranged from 23.4% to 50.6%; the biomechanical analyses did not quantify negative work.Source: §V · p. 7

VI. CONCLUSION

The conclusion is that assistance delay affects mechanical and metabolic outcomes differently, with benefits concentrated in level-ground and ramp-ascent walking rather than reliably extending to ramp descent. Mapping evaluation should reflect the intended outcome and the assisted joint’s mechanical role.

VI. CONCLUSION

  • Delay affected mechanical unloading more than metabolic benefit, and benefits did not consistently extend to RD, suggesting mechanically distinct tasks may require mapping changes beyond delay.Shared mappings may generalize across tasks with overlapping joint-level mechanics.Source: §VI · p. 7
Loading 2609.25417v1…