PaperlayerView exact revision on arXiv

Source-linked AI summary

A Transparent 3-DoF Assistive Device for Upper-Limb Tremor and Ataxia Compensation: Design and Experimental Validation

Thomas Dobosz, Veronika Hofmann, Ahmed Barhoma, Ann-Kathrin Henke, Johannes A. Gaus, Hans Mees, Winfried Ilg, Isabell Wochner, Martin Giese, Andreas Bulling, Syn Schmitt, Christophe Maufroy, Urs Schneider, Thomas Bauernhansl, Daniel Haeufle

arXiv:2610.30955v1cs.RO

TL;DR

Assistive devices for neurodegenerative movement disorders must combine transparency with corrective force rendering, a difficult trade-off when resistance can impede voluntary motion. This paper presents and evaluates a 3-DoF platform using a hybrid control strategy. Experiments show transparent horizontal behavior and force rendering, while vertical-axis friction limits performance.

  • Problem

    Assistive devices must combine high mechanical transparency with corrective forces, while friction, inertia, and resistance in force-capable systems can interfere with voluntary motion.

    Source: §I · p. 1

  • Method

    The paper presents and experimentally evaluates a completed 3-DoF assistive platform with a hybrid force-control strategy.

    Source: §I · p. 2

  • Results

    The platform provided approximately 10 N of omnidirectional force capability and accurate horizontal force rendering across the tested frequency range.

    Source: §IV · p. 6

  • Takeaways & Limitations

    The results establish iAssistADL as a promising basis for future tremor and ataxia compensation during activities of daily living.

    Source: §V · p. 7

    Force-rendering tests used a rigid instrumented testbed and only three representative ADL configurations, with dynamic evaluation limited to 5 Hz.

    Source: A. Limitations · p. 7

Abstract

from arXiv · show

Assistive devices for patients with neurodegenerative movement disorders must combine high mechanical transparency with the ability to generate corrective interaction forces. Achieving both properties simultaneously remains challenging. This paper presents the completed iAssistADL platform, a 3-DoF assistive device for future tremor and ataxia compensation during activities of daily living. The system employs a transparency-oriented mechanical architecture with all actuators located at the base, resulting in low moving mass, decoupled kinematics, and high mechanical transparency. To compensate for friction-induced transparency limitations along the vertical axis, a hybrid control strategy is proposed that combines open-loop force rendering in the horizontal plane with closed-loop force control in the vertical direction. The platform is experimentally evaluated through transparency measurements and static and dynamic force-rendering experiments at representative configurations. Results demonstrate resistive torques below 0.1 Nm in the horizontal plane, end-effector forces of up to approximately 10 N throughout the workspace, and accurate horizontal force rendering up to 4 Hz. While the vertical axis remains limited by friction-induced dynamics at higher frequencies, the proposed system successfully combines high transparency with controlled force generation, establishing a promising hardware platform for future tremor and ataxia compensation.

I. INTRODUCTION

Neurodegenerative movement disorders can disrupt essential daily activities even when voluntary movement is preserved, creating a need for assistance that counters pathological motion without impeding intended movement. This work presents and experimentally evaluates a 3-DoF platform combining mechanical transparency with force rendering.

I. INTRODUCTION

  • Patients with Parkinson’s disease or cerebellar ataxia may have impaired daily activities because tremor or dysmetric overshooting interferes with movement despite largely preserved voluntary motion,,,,.Assistive technologies should selectively compensate pathological motion while preserving voluntary motor control [6].Source: §I · p. 1
  • High transparency and sufficient force rendering remain challenging to combine because force-capable devices can add friction, inertia, and resistance that interfere with voluntary motion.Related end-effector robots address low mechanical impedance and active force rendering, but target rehabilitation and motor-control applications rather than pathological-movement compensation,.Source: §I · p. 1
  • Damping can slow oscillations but cannot direct an overshooting reach toward its intended endpoint, whereas actively rendered forces can provide that directional correction.Added limb mass improved some single-joint reaches but did not carry over to multi-joint movement.Source: §I · p. 2
  • The completed 3-DoF iAssistADL platform adds a hybrid force-control strategy and experimental force-rendering validation to the system concept introduced in prior work.It was evaluated through transparency measurements and force-rendering experiments at representative ADL configurations for future compensation of pathological movement components.Source: §I · p. 2

II. METHODS

  • The platform uses open-loop force generation in the horizontal plane and closed-loop force control vertically to address friction observed on the z-axis.It was designed for forces up to 10 N at the forearm and evaluated at positions corresponding to grasping, pouring, and drinking.Source: §II · p. 2

A. System Design

The 3-DoF design places actuation on the fixed base and uses remote transmissions to limit moving mass. Its kinematics separate vertical motion from the horizontal plane, supporting axis-wise control.

1) Mechanical Architecture and Actuation Concept:

  • All actuators are mounted on the fixed base and connected through remote transmissions to reduce moving mass and apparent inertia.The two rotational joints provide horizontal motion, while a prismatic joint provides vertical motion.Source: 1) Mechanical Architecture and… · p. 2
  • A ball-spline transmits torque through the vertically translating stage, while a cable-driven Capstan integrated into the scissor mechanism actuates the distal rotational joint.Routing cables through the scissor linkage maintains nearly constant cable lengths across the workspace; a passive spring mechanism balances arm weight, and a forearm interface with three passive rotational DoFs permits natural arm motion while limiting parasitic moments.Source: 1) Mechanical Architecture and… · pp. 2–3, Figure 3 · p. 2

2) Kinematic Model:

  • The two planar links each have length 0.45 m, and the vertical stage has effective length 0.30 m.The planar joint angles are measured in the device base frame, while θ3 actuates the vertical stage.Source: 2) Kinematic Model · p. 3
  • The Jacobian’s block-diagonal structure decouples vertical motion from the horizontal plane, allowing closed-loop z-axis force control without cross-talk into the open-loop x–y mapping.The inverse kinematics also provide a closed-form solution; the sign of ∆ selects the elbow branch, and the physical build uses θ1 < θ2.Source: 2) Kinematic Model · p. 3

B. Transparency-Oriented Control Concept

The control objective is to preserve voluntary movement when no assistive force is commanded while superimposing controlled end-effector assistance. The axis-wise strategy uses open-loop control where feasible and closed-loop control where friction makes it necessary.

B. Transparency-Oriented Control Concept

  • With no commanded assistance, the device should not impede voluntary motion; assistance is added as a controlled Cartesian end-effector force.The axis-wise strategy follows the principle “open-loop where possible, closed-loop where necessary.”Source: B. Transparency-Oriented Control Concept · p. 3
  • In the horizontal plane, backdrivability and a near-linear, current-inferable torque–force relation allow open-loop force delivery through the Jacobian-transpose map without a dedicated planar sensor.Project investigations indicated that approximately 5 N could compensate relevant pathological movement components during representative ADLs.Source: B. Transparency-Oriented Control Concept · p. 3
  • The z-axis uses closed-loop force control with a uniaxial load cell because its approximately 3 N dry-friction load is on the order of the desired assistive force.The partition also limits sensing hardware to a single force sensor, consistent with the concept proposed in.Source: B. Transparency-Oriented Control Concept · p. 3

1) Open-Loop Control in the Horizontal Plane:

The horizontal-plane force command is mapped to motor torque using the Jacobian transpose and gear ratios. The deployed implementation assumes unit transmission efficiency, so mechanical losses cause systematic force under-delivery.

1) Open-Loop Control in the Horizontal Plane:

  • The Jacobian-transpose map converts the desired force into motor-torque commands using the gear ratios.The end-to-end accuracy of this mapping is assessed through force-tracking results rather than a reference joint-torque measurement.Source: 1) Open-Loop Control in the Horizontal… · p. 3
  • The deployed code sets η_i = 1, while experimentally estimated efficiencies of 0.92–0.96 imply systematic under-delivery from mechanical losses.Efficiencies were estimated from static force-transmission measurements comparing commanded motor torques with measured end-effector forces.Source: 1) Open-Loop Control in the Horizontal… · p. 3
  • The physically realized capability envelope uses measured transmission efficiencies and is therefore more conservative than the code-nominal command.The map is evaluated end-to-end through force tracking.Source: 1) Open-Loop Control in the Horizontal… · p. 3

2) Gravity and Friction Compensation in z:

The z-axis combines feedforward compensation for gravity and friction with load-cell-based feedback to regulate residual force error. The same regulator renders zero force or a commanded assistive force, while its tuning prioritizes low perceived resistance.

2) Gravity and Friction Compensation in z:

  • The z-axis feedforward compensates predictable gravity and friction disturbances, while a passive counterbalance mechanically offloads the quasi-static gravitational load.No gravity term is included in the control law, preserving actuator torque headroom.Source: 2) Gravity and Friction Compensation… · p. 3

3) Closed-Loop Force Control in the Vertical Direction:

  • Ball-spline friction and stick–slip prevent accurate z-axis force rendering with feedforward alone, so a PI controller regulates residual force error using load-cell measurements.Its output is realized as motor torque through the z-row mapping; the integral action rejects the true transmission loss.Source: 3) Closed-Loop Force Control in the… · p. 3
  • At a zero force command, the regulator cancels friction and drives residual interaction force to zero; nonzero commands are tracked by the same regulator.This allows assistance to be superimposed on force-neutral behavior.Source: 3) Closed-Loop Force Control in the… · p. 3
  • The controller was tuned to minimize perceived resistance while maintaining low steady-state force errors during user interaction.The reported gains were manually tuned on the identified z-axis plant.Source: 3) Closed-Loop Force Control in the… · p. 4

4) Force Controller Design:

The force controller uses a PI compensator and interaction-oriented manual tuning. The derivative term is omitted to avoid amplifying load-cell noise in the tremor band.

4) Force Controller Design:

  • The controller uses a PI compensator, with its derivative term omitted to avoid amplifying load-cell noise in the tremor band.The control loop runs at 1 kHz (Ts = 1 ms).Source: 4) Force Controller Design · p. 4
  • Kp = 0.1 and Ki = 0.2 were manually tuned on the identified z-axis plant by suppressing perceived friction without overcorrecting and injecting resistance.The gains were increased until the device followed the user freely, but kept below the level that caused overcorrection.Source: 4) Force Controller Design · p. 4
  • The interaction-oriented tuning yielded smoother, less intrusive behavior than Ziegler–Nichols tuning of the same loop, which produced an over-aggressive, oscillatory response.The authors selected felt behavior as more representative of the transparency objective than a model-based rule optimized for reference tracking.Source: 4) Force Controller Design · p. 4

C. Experimental Validation

The validation measures mechanical transparency and force-rendering performance at representative ADL configurations. Results support open-loop force delivery in the horizontal plane and motivate closed-loop control on the friction-affected vertical axis.

C. Experimental Validation

  • The validation assesses horizontal-plane transparency as the basis for open-loop operation and evaluates closed-loop z-axis force accuracy across the ADL workspace.It characterizes per-axis friction, then tests force-rendering accuracy, bandwidth, and practical limits under static and dynamic loading.Source: C. Experimental Validation · p. 4

1) Transparency Evaluation:

  • The rotational axes exhibited resistive torques below 0.1 N m, while the vertical axis had approximately 3 N of static friction, motivating the hybrid control partition.The partition combines open-loop force delivery in the horizontal plane with dedicated force-feedback control on z.Source: 1) Transparency Evaluation · p. 4

2) Force-Rendering Evaluation:

  • Force commands were tested at three ADL operating points—grasp, pour, and drink—using an external force/torque sensor as an independent ground-truth reference.Tests applied individual Cartesian-axis commands and a diagonal planar command, with steps, sinusoids, and frequency sweeps up to 5 Hz.Source: 2) Force-Rendering Evaluation · pp. 4–5

III. RESULTS

The capability analysis establishes a workspace-wide lower bound for planar force and reports pose-dependent vertical and maximum planar capabilities. The tested 2–5 N range lies inside the guaranteed planar envelope.

A. Force Capability

  • 9.89 N is the guaranteed omnidirectional planar force throughout the reachable workspace, fixed by the shoulder-joint torque limit and the tighter half-plane constraint.The 2–5 N test range lies inside this conservative envelope, so the experiments probe rendering fidelity rather than saturation.Source: A. Force Capability · p. 5
  • 10.0–10.5 N is the pose-dependent vertical capability, while maximum planar force ranges from 15.0 N at P2 to 28.8 N at P3.These are pose-dependent capabilities reported alongside the workspace-wide guaranteed planar force.Source: A. Force Capability · p. 5
  • Planar force anisotropy rises from 1.51 to 2.91 as manipulability det Jxy degrades toward P3.Anisotropy is the ratio of maximum planar force to guaranteed omnidirectional planar force.Source: A. Force Capability · p. 5

B. Transparency

The transparency measurements support open-loop force rendering in the horizontal plane and closed-loop control vertically, where friction limits performance. Across the tested band, horizontal tracking remains accurate, while vertical performance degrades above about 2 Hz.

B. Transparency

  • Horizontal rotational joints have average resistive torques below 0.1 Nm, while the vertical prismatic axis has higher resistance from the ball-spline mechanism.The device reduced reachable workspace by only 2.2%; these results motivated open-loop horizontal rendering and dedicated vertical force feedback.Source: B. Transparency · p. 5

C. Static Force Rendering (Step Response)

  • The horizontal open-loop map reaches the commanded step level within approximately 16 ms and settles in approximately 1.3 s with 2–5% steady-state error.The vertical axis has comparable mean steady-state accuracy but slower, more variable transients: 3.0 s rise time, 7.0 s settling, and σess ≈ 3 N.Source: C. Static Force Rendering (Step… · p. 5

D. Dynamic Force Tracking

  • The overall transport delay is approximately 0.13–0.19 s, producing phase lag that increases with frequency and dominating raw RMSE of 3–5 N.An average delay of approximately 98 ms between motor motion and end-effector force indicates a substantial mechanical-transmission contribution.Source: D. Dynamic Force Tracking · p. 5
  • Horizontal amplitude tracking is preserved up to 4 Hz, with amplitude ratio 0.91–0.97 and phase-compensated RMSE 0.36–0.68 N.The vertical loop tracks at 2 Hz (Ar = 0.98) but amplifies at 4 Hz (Ar = 1.87, phase-compensated RMSE 3.4 N).Source: D. Dynamic Force Tracking · p. 5

E. Frequency-Sweep and Operating Limits

  • Horizontal axes maintain high tracking quality across 0.05–5 Hz, while the vertical mean magnitude ratio reaches 1.40 and peaks at 2.7 near 3.2 Hz.Mean coherence is 0.97/0.95 for x/y and mean magnitude ratio is 1.01/0.96; sweeps stopped at visible oscillation onset.Source: E. Frequency-Sweep and Operating Limits · pp. 5–6
  • The horizontal open-loop channel remains accurate across the tested band, while the vertical closed-loop channel resonates above about 2 Hz and does not meet the upper tremor-band target.The platform meets its assistive ADL force-rendering target, but the present z-axis tuning does not satisfy frequencies above 2 Hz.Source: E. Frequency-Sweep and Operating Limits · p. 6

IV. DISCUSSION

The results largely achieve the platform’s objective of combining mechanical transparency with force generation. Horizontal performance supports table-based ADLs and possible low-frequency compensation, while vertical friction and transport delay constrain performance.

IV. DISCUSSION

  • The platform largely achieves its objective of combining high mechanical transparency with effective force generation, with approximately 10 N omnidirectional capability and accurate horizontal rendering across the tested frequencies.The authors identify table-based activities such as drinking, pouring, and object manipulation as suitable target scenarios.Source: §IV · p. 6
  • Horizontal transparency and force rendering support table-based ADLs, while the measured capability suggests a basis for future compensation of ataxic reaching and low-frequency tremor components.The paper notes that corrective forces for ataxia are primarily needed at relatively low frequencies during goal-directed reaches.Source: §IV · p. 6
  • Vertical ball-spline friction and its interaction with lightweight-arm compliance limit closed-loop performance, causing longer step settling and resonant amplification above approximately 2 Hz.The authors identify refined friction compensation and improved moving-structure stiffness-to-mass ratio as needed for vertical-axis improvement.Source: §IV · pp. 6–7, Figure 6 · p. 6
  • The approximately 140 ms transport delay is mainly mechanical, with roughly 98 ms arising between motor actuation and end-effector force generation.Offline compensation can address the delay when characterizing performance, but it may affect future real-time tremor suppression.Source: §IV · p. 7, Figure 7 · p. 7

A. Limitations

The evaluation is limited to rigid-testbed experiments at representative configurations and frequencies, with no human participants; the conclusion identifies vertical-axis stick–slip as the main bandwidth constraint.

A. Limitations

  • Force-rendering tests used a rigid instrumented testbed, so human-arm impedance could alter observed stick–slip behavior and closed-loop response.The study did not test direct interaction with a human user.Source: A. Limitations · p. 7
  • Evaluation covered only three representative ADL configurations and excitation frequencies up to 5 Hz; higher-frequency behavior requires further validation.Limited trials and considerable vertical-axis variability also produced wider confidence intervals for z-axis performance metrics.Source: A. Limitations · p. 7

V. CONCLUSION

  • Vertical-axis friction-induced stick–slip currently restricts the achievable bandwidth, and evaluation in dynamic human-in-the-loop experiments and patient studies remains future work.The reported experiments were conducted exclusively on a rigid laboratory testbed without human participants.Source: §V · p. 7, ETHICS STATEMENT · p. 7

ETHICS STATEMENT

  • The experiments were conducted on a rigid laboratory testbed without human participants; the cited ethics approval concerns other project studies.The transparency ratings cited from were obtained under the corresponding approval procedures.Source: ETHICS STATEMENT · p. 7
Loading 2610.30955v1…