Embedded Files
Keywords: Research Project, Prostheses, Mode Classification, Gait Phase Estimation, Sliding Window Algorithm
Advisor: Prof. Patrick M. Wensing (University of Notre Dame)
Collaborator: Dr. Ryan Posh
Recognizing and identifying human locomotion is a critical step to ensuring fluent control of wearable robots, such as transtibial prostheses. In particular, classifying the locomotion mode and estimating the gait phase are key. In this work, a novel, interpretable, and computationally efficient algorithm is presented for simultaneously predicting locomotion mode and gait phase. Using able-bodied (AB) data and transtibial prosthesis (PR) data collected via a bypass adapter, seven locomotion modes are tested including slow, medium, and fast level walking (0.6, 0.8, and 1.0 m/s), ramp ascent/descent (5 degrees), and stair ascent/descent (20 cm height). Overall classification accuracy was 99.1% and 99.3% for the AB and PR conditions, respectively. The average gait phase error across all data was less than 4%. Exploiting the structure of the data, computational efficiency reached 2.91 us per time step. The time complexity of this algorithm scales as O(NM) with the number of locomotion modes M and samples per gait cycle N. This efficiency and high accuracy could accommodate a much larger set of locomotion modes (~ 700 on the Open-Source Leg Prosthesis) to handle the wide range of activities pursued by individuals during daily living.
Illustration of the sliding kernel trick
Illustration of the sliding kernel trick
The sliding kernel matrix K^m is compared to historical data D^m via the sum of squares error (SSE). All calculations can be repeated for each time step, such as time = i (top) and time = i+1 (middle), or computation can be significantly reduced by exploiting the many shared computations between subsequent time steps (bottom).
Classification and phase estimation result for one experiment
Classification and phase estimation result for one experiment
The minimum sum-of-squares error e_i^m of all activities (top) are compared, with the lowest error used to predict the current locomotion mode (middle) and the current gait phase (bottom).
Benchmark results for the proposed method on two different platforms
Benchmark results for the proposed method on two different platforms
The proposed computation reduction approach leads to a significant improvement in computation time, being 310 or 1000 times faster compared to the naive method on the target device or MacBook.
Google Sites
Report abuse