Human-Exoskeleton Kinematic Calibration to Improve Hand Tracking for Dexterous Teleoperation

TL;DR

This work tackles accurate hand tracking in exoskeleton-based teleoperation by introducing a subject-specific calibration that learns virtual link parameters via residual-weighted optimization on a four-bar closed-loop kinematic model. Implemented on the Maestro exoskeleton with seven participants, the approach uses a two-phase calibration protocol and data-driven weight refinement to align device measurements with anatomical joints, validated through motion capture and Unity visualizations. Quantitatively, calibration reduces joint and fingertip errors, with index fingertip accuracy improving by approximately 71% on average and mean fingertip error around 10 mm, and qualitative visuals confirm enhanced motion fidelity. The framework generalizes to other sensorized hand exoskeletons and supports high-fidelity teleoperation and robot learning with minimal external sensing.

Abstract

Hand exoskeletons are critical tools for dexterous teleoperation and immersive manipulation interfaces, but achieving accurate hand tracking remains a challenge due to user-specific anatomical variability and donning inconsistencies. These issues lead to kinematic misalignments that degrade tracking performance and limit applicability in precision tasks. We propose a subject-specific calibration framework for exoskeleton-based hand tracking that estimates virtual link parameters through residual-weighted optimization. A data-driven approach is introduced to empirically tune cost function weights using motion capture ground truth, enabling accurate and consistent calibration across users. Implemented on the Maestro hand exoskeleton with seven healthy participants, the method achieved substantial reductions in joint and fingertip tracking errors across diverse hand geometries. Qualitative visualizations using a Unity-based virtual hand further demonstrate improved motion fidelity. The proposed framework generalizes to exoskeletons with closed-loop kinematics and minimal sensing, laying the foundation for high-fidelity teleoperation and robot learning applications.

Figures (10)

• Figure 1: (Left) The Maestro hand exoskeleton worn on the thumb, index, and middle fingers. (Right) Simplified kinematic model of the exoskeleton and corresponding rendering of the mapped virtual hand.
• Figure 2: Kinematic model of the Maestro finger–exoskeleton four-bar interface. The thumb is equipped with redundant joints $(\delta_1,\delta_2')$ and is modeled with three kinematic loops: one RRPR loop and two RRRR loops. The index and middle fingers are equipped with redundant joints $(\delta_1,\delta_2)$ and are modeled with two kinematic loops: one RRPR loop and one RRRR loop. The $X$–$Y$ frame is defined such that the $X$-axis is aligned longitudinally with the base frame and the $Y$-axis is perpendicular to it. All variable definitions are provided in Tab. \ref{['tab:equations']}.
• Figure 3: Sensitivity analysis in simulation of fingertip position error (mm) as a function of perturbations in select parameters (% variation).
• Figure 4: Visualization of the kinematic parameter calibration with weighted optimization. The human user performs a two-phase calibration, and the weight distribution is adjusted based on human data.
• Figure 5: Experimental setup and workflow. (a) Motion capture markers placement on the thumb, (b) on the index finger, and (c) experiment workflow for human subject testing.