Pitts MN, Ebers MR, Agresta CE, Steele KM (2025) Evaluating Sparse Inertial Measurement Unit Configurations for Inferring Treadmill Running Motion

Journal Article in Sensors

 Inertial measurement units (IMUs) are used to analyze running performance. While leveraging one sensor to estimate kinematic and kinetic variables is common, sparsity limits the number of digital biomarkers that can be evaluated.

An illustration demonstrating experimental factors influencing the accuracy of motion inference when using Shallow recurrent decoder networks (SHRED) can reconstruct a dense set of time-series signals from a single input sensor. Running speed and sampling rate were most influential, while sensor location, and sensor type were neutral.Aim: Shallow recurrent decoder networks (SHRED) can reconstruct a dense set of time-series signals from a single input sensor and have been successful in human mobility applications, highlighting the potential for this algorithm to monitor running.

Methods: We trained and tested subject-specific SHRED models of nine subjects running on a treadmill to map from one input sensor to the remaining three IMUs. We varied the type of input to reflect experimental parameters that are important in running studies—sensor location, sensor type, sampling rate, and running speed—and compared the error of inferred signals from each input type.

Results: Sensor location and type did not impact SHRED inference accuracy, while decreasing the sampling rate affected the accuracy of ankle measurements. All ankle acceleration inferences from these models remained below the minimal detectable change threshold of 12.0 m/s2. SHRED models trained and tested at multiple speeds did not accurately infer IMU measurements below this threshold.

Interpretation: SHRED may broaden the scope of motion analysis by expanding access to data with fewer sensors. The data from this study and an instructional Jupyter notebook for training and testing individualized SHRED models are available at [link to GitHub].

ASB 2024 Recap

Steele Lab members, Charlotte Caskey, Victoria (Tori) Landrum, and Megan Ebers, attended the American Society of Biomechanics Annual Meeting (ASB) in Madison, WI from August 5-8, 2024.

Charlotte gave a poster presentation on the “Effect of spinal stimulation and interval treadmill training on gait mechanics in children with cerebral palsy”

Tori also gave a poster presentation on the “Impact of a Resistive Exoskeleton on Fatigue in Children with Cerebral Palsy”

Megan co-hosted a Symposia Session titled, “Can machine learning reveal the next generation of neural and biomechanical processes governing human movement?” with Steele Lab Alumni, Michael Rosenberg. In Megan’s talk, “A machine learning approach to quantify individual gait responses to ankle exoskeletons,” she discussed how neural network-based discrepancy modeling can be used to isolate the dynamics governing changes in gait with ankle exoskeletons.

CMBBE 2024 Recap

Members of the Steele Lab traveled to Vancouver, BC for the 19th International Symposium on Computer Methods in Biomechanics and Biomedical Engineering (CMBBE) hosted by the University of British Columbia.

PhD candidate, Mackenzie Pitts, gave a poster presentation on “Inferring Unmeasured Inertial Data from Sparse Sensing for Treadmill Running”. Steele Lab Alumni and Post-Doctoral Research Fellow at Emory University, Michael Rosenberg, gave a podium presentation titled “Recurrent Neural Network Gait Signatures Encode Speed-Induced Changes in Post-Stroke Gait Quality.”

In addition to sharing their research at the conference, the Steele Lab enjoyed connecting with fellow biomechanics and biomedical engineering researchers as well as exploring the beautiful campus at UBC.

MR Ebers, JP Williams, KM Steele, JN Kutz (2024) “Leveraging arbitrary mobile sensor trajectories with shallow recurrent decoder networks for full-state reconstruction,”

Journal Article in IEEE Access

Sensing is a fundamental task for the monitoring, forecasting, and control of complex systems. In many applications, a limited number of sensors are available and must move with the dynamics. Currently, optimal path planning, like Kalman filter estimation, is required to enable sparse mobile sensing for state estimation. However, we show that arbitrary mobile sensor trajectories can be used. By adapting the Shallow REcurrent Decoder (SHRED) network with mobile sensors, their time-history can be used to encode global information of the measured high-dimensional state space.

Summary figure of a shallow recurrent decoder network (SHRED) leveraging mobile sensors to reconstruct full state-space estimates from sparse dynamical trajectories. (Left) Sensor trajectory history encodes global information of the spatio-temporal dynamics of the sparsely measured system. In this work, we evaluate three challenging datasets, including forced isotropic turbulence, global sea-surface temperature, and human biomechanics. (Middle) The mobile SHRED architecture can (i) embed the multiscale physics of a system into a compact and low-dimensional latent space, and (ii) provide a mapping from the sparse mobile sensors to a full state estimate. (Right) The high-dimensional and complex system states can be reconstructed, provided training data for the dynamical trajectory of the sensor(s) is available.Aim: We leverage sparse mobile sensor trajectories for full-state estimation, agnostic to sensor path.

Methods: Using modern deep learning architectures, we show that a sequence-to-vector model, such as an LSTM (long, short-term memory) network, with a decoder network, dynamic trajectory information can be mapped to full state-space estimates.

Results: We demonstrate that by leveraging mobile sensor trajectories with shallow recurrent decoder networks, we can train the network (i) to accurately reconstruct the full state space using arbitrary dynamical trajectories of the sensors, (ii) the architecture reduces the variance of the mean-squared error of the reconstruction error in comparison with immobile sensors, and (iii) the architecture also allows for rapid generalization (parameterization of dynamics) for data outside the training set. Moreover, the path of the sensor can be chosen arbitrarily, provided training data for the spatial trajectory of the sensor is available.

Interpretation: The time-history of mobile sensors can be used to encode global information of the measured high-dimensional state space.

Introducing Dr. Charlotte DeVol Caskey!

Congratulations to Dr. Charlotte DeVol Caskey on earning her Doctorate in Mechanical Engineering! Dr. Caskey’s PhD thesis dissertation was titled Effects of Spinal Stimulation on Neuromechanics of Gait for Children with Cerebral Palsy. Congratulations and best of luck as you move forward as a Postdoc in the Human Neuromechanics Laboratory at the University of Florida in Gainesville!