Robust Plug-and-Play Joint Axis Estimation Using Inertial Sensors

Fredrik Olsson, Manon Kok, Thomas Seel, Kjartan HalvorsenSensors 2020 · OpenReview

TL;DR

This paper introduces a "plug-and-play" method for calibrating inertial sensors on hinge joints, eliminating the need for specific user movements. It achieves this by intelligently selecting informative data from arbitrary motions and quantifying the uncertainty of its estimates, ensuring reliable joint axis identification.

What Problem Does This Paper Address?

Accurate motion capture using wearable inertial measurement units (IMUs) relies on precisely aligning sensors with body segments. For hinge joints (like knees or fingers), this means identifying the joint's axis in the sensor's coordinate system. Existing methods often require users to perform specific, informative movements within a defined time window, which is inconvenient and doesn't guarantee success. Furthermore, users have no way of knowing if the calibration was actually accurate. This paper tackles the challenge of creating a truly "plug-and-play" system that works with any motion, automatically identifies the joint axis, and informs the user about the calibration's reliability.

Methods

The core of the proposed method lies in three key innovations:

  • Intelligent Sample Selection: Instead of relying on a dedicated calibration phase, the system can use data from any motion. It employs algorithms to select only the most informative samples from a larger dataset, discarding redundant or non-informative ones. This is achieved by analyzing angular velocity differences between sensors (for gyroscope data) and by penalizing stationary sensor data (for accelerometer data).
  • Combined Sensor Fusion: The method leverages both accelerometer and gyroscope data. It combines constraints derived from both sensor types and weights them to achieve a globally optimal estimate of the joint axis. This fusion makes the estimation robust to different motion types.
  • Uncertainty Quantification: A crucial aspect is providing the user with confidence in the calibration. The paper introduces a method to quantify both local (based on the optimization's Jacobian) and global (based on multiple initializations and sequential estimates) uncertainty. This allows the system to indicate when the calibration is sufficiently accurate.

The joint axis is parameterized using spherical coordinates, and an optimization problem is formulated to minimize the residuals from both angular velocity and acceleration constraints. A Gauss-Newton method is used to solve this optimization problem, with a re-initialization step to ensure the correct sign pairing of the joint axis is found.

Datasets

The method was validated using data collected from a 3D printed hinge joint system equipped with two wireless IMUs. This system performed 14 different motions, including stationary states, free rotations, sequential rotations, simultaneous planar rotations, and combinations at different speeds. These motions were categorized into four scenarios to test the method's robustness to the order and type of data. No new datasets were introduced; existing experimental setups were used.

Key Insights and Ideas

  • Arbitrary Motion is Sufficient: The research demonstrates that informative data for joint axis estimation can be found in almost any motion, as long as the joint isn't completely stiff. This significantly reduces the burden on the user.
  • Data Quality Over Quantity: By intelligently selecting a subset of data, the method can achieve comparable or even better results than using all available data, especially when the data contains many non-informative segments. This also reduces computational load.
  • Knowing When You're Done: The uncertainty quantification is a game-changer. It provides a clear indication of when the calibration is reliable, preventing users from relying on potentially inaccurate results.
  • Robustness to Noise and Bias: The method shows good performance even with added artificial sensor biases, highlighting its practical applicability.

Key Results

  • The proposed method achieved angular errors in the order of 2° for a mechanical joint.
  • This accuracy was obtained using a selected subset of 125–1000 samples, significantly less than the total available data.
  • The method demonstrated robustness to sensor biases, with errors remaining relatively low even with added artificial biases of 1 m/s² and 1°/s.
  • The uncertainty quantification successfully identified estimates that met acceptable accuracy criteria (Emax = 3°) 100% of the time when using a sufficient number of sequential estimates (nmin = 10).

Limitations and Future Work

  • Non-Rigid Systems: The authors acknowledge that the method's reliability in non-rigid systems, such as human limbs with soft tissue motion, requires further investigation.
  • Bias Compensation: While the method is robust to bias, pre-calibration is still recommended. For significant bias drift, online compensation methods might be needed in conjunction with this approach.
  • Extension to Other Joint Types: The current method is designed for hinge joints (1-DOF). Future work could extend it to other joint types (e.g., 2-DOF or 3-DOF joints).
  • Real-time Applications: While the sample selection algorithms are designed for online implementation, further exploration of real-time performance and integration into assistive devices is suggested.

Summary updated

Abstract

Inertial motion capture relies on accurate sensor-to-segment calibration. When two segments are connected by a hinge joint, for example in human knee or finger joints as well as in many robotic limbs, then the joint axis vector must be identified in the intrinsic sensor coordinate systems. Methods for estimating the joint axis using accelerations and angular rates of arbitrary motion have been proposed, but the user must perform sufficiently informative motion in a predefined initial time window to accomplish complete identifiability. Another drawback of state of the art methods is that the user has no way of knowing if the calibration was successful or not. To achieve plug-and-play calibration, it is therefore important that 1) sufficiently informative data can be extracted even if large portions of the data set consist of non-informative motions, and 2) the user knows when the calibration has reached a sufficient level of accuracy. In the current paper, we propose a novel method that achieves both of these goals. The method combines acceleration- and angular rate information and finds a globally optimal estimate of the joint axis. Methods for sample selection, that overcome the limitation of a dedicated initial calibration time window, are proposed. The sample selection allows estimation to be performed using only a small subset of samples from a larger data set as it deselects non-informative and redundant measurements. Finally, an uncertainty quantification method that assures validity of the estimated joint axis parameters, is proposed. Experimental validation of the method is provided using a mechanical joint performing a large range of motions. Angular errors in the order of 2 ∘ were achieved using 125–1000 selected samples. The proposed method is the first truly plug-and-play method that overcome the need for a specific calibration phase and, regardless of the user’s motions, it provides an accurate estimate of the joint axis as soon as possible.
BibTeX
@inproceedings{olsson2020robust,
  author = {Fredrik Olsson and Manon Kok and Thomas Seel and Kjartan Halvorsen},
  title = {{Robust Plug-and-Play Joint Axis Estimation Using Inertial Sensors}},
  booktitle = {Sensors},
  year = {2020},
  url = {https://openreview.net/forum?id=GB11ugyrAIj},
}
Figure