Simultaneous Locomotion Mode Classification and Continuous Gait Phase Estimation for Transtibial Prostheses

TL;DR

A novel, interpretable, and computationally efficient algorithm is presented for simultaneously predicting locomotion mode and gait phase and could accommodate a much larger set of locomotion modes to handle the wide range of activities pursued by individuals during daily living.

Abstract

Recognizing and identifying human locomotion is a critical step to ensuring fluent control of wearable robots, such as transtibial prostheses. In particular, classifying the intended 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) and transtibial prosthesis (PR) data, 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 $μ$s per time step. The time complexity of this algorithm scales as $O(N\cdot M)$ 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 ($\sim$ 700 on Open-Source Leg Prosthesis) to handle the wide range of activities pursued by individuals during daily living.

Figures (7)

• Figure 1: The accuracy and number of locomotion modes present in representative locomotion mode classification literature. Studies in red present classification only and studies in blue classify the locomotion mode and estimate the gait phase.
• Figure 2: Experimental data was collected from 7 different locomotion modes: slow/medium/fast walking, ramp ascent/descent, and stair ascent/descent), with 2 different conditions: able-bodied (AB) and with the Open-Source Leg (OSL) transtibial prostheses (PR).
• Figure 3: A window of historical data is compared to sliding locomotion mode kernels. Locomotion modes include slow/medium/fast walking (Slow, Med, Fast), ramp ascent/descent (RA, RD), and stair ascent/descent (SA, SD). The kernel with the lowest sum of squares error (SSE) is selected as the current locomotion mode (shown as SA, SSE = 2), and the sliding configuration that results in that minimum SSE determines the gait phase (shown as $\phi = 30\%$).
• Figure 4: 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 = t (left) and time = t+1 (right), or computation can be significantly reduced by exploiting the many shared computations between subsequent time steps (bottom).
• Figure 5: Confusion matrices summarizing the steady-state classification results for the able-bodied (top) and transtibial prosthesis (bottom) conditions. Slow, Med, Fast, RA, RD, SA, and SD correspond to steady state locomotion modes of 0.6, 0.8, and 1.0 m/s level walking, ramp ascent/descent, and stair ascent/descent, respectively.