Mode-Unified Intent Estimation of a Robotic Prosthesis Using Deep-Learning

Traditional robotic knee-ankle prostheses categorize ambulation modes such as level walking, ramps, and stairs. However, human movement scales continuously across various states rather than discretely, making traditional mode classifiers inadequate for accurate intent recognition. This letter proposes a mode-unified intent recognition strategy that continuously estimates terrain slopes across five modes: level ground, ramp ascent/descent, and stair ascent/descent. Locomotion data from 16 individuals with transfemoral amputation were utilized to train slope estimation and mode classification models based on deep temporal convolutional networks. The proposed method was compared to the traditional mode classifier via offline test, using leave-one-subject-out validations for the user-independent performance. The mode-unified slope estimator achieved an MAE of 1.68 ± 0.60 degrees, outperforming the mode classifier's MAE of 1.94 ± 0.97 degrees (p<0.05). The lower slope estimation errors resulted in higher accuracy in replicating knee kinematics of able-bodied subjects, with the proposed system achieving an average MAE of 5.13 ± 2.00 degrees for knee clearance and 6.74 ± 2.97 degrees for knee contact angle, compared to the traditional classifier's 12.10 ± 5.20 degrees and 13.80 ± 3.28 degrees (p<0.01), respectively, in stair ascent. These results suggest that our mode-unified approach can enable continuous adjustment to terrains without mode classification.

I. Introduction

Robotic knee-ankle prostheses are developed to enable individuals with transfemoral amputations to navigate various walking environments they face daily. As community ambulation involves different walking modes, such as level walking, ramps, and stairs, recent prosthesis studies have been dedicated to generating lower limb dynamics that adjust to environments [1]. Conventional prosthesis systems recognize terrain conditions (i.e., ambulation mode) to switch to the corresponding mode-specific controller [2], [3], [4], [5]. However, as people take thousands of strides in various modes daily [3], a prosthesis user would encounter dozens of potential falls due to kinematic and kinetic mismatches induced by incorrect mode switches, even with a 99%-accurate mode classifier. Furthermore, Human dynamics vary not only across ambulation modes, but also within each mode [6], [7]. Thus, the traditional methods categorizing user intent as discrete modes cannot account for within-mode variations in prosthesis systems. Recent studies have addressed these issues by implementing continuously varying control parameters within specific subsets of modes, such as level ground and ramps [8], [9], [10] or level ground and stairs [11], [12]. However, these works were conducted in limited environmental combinations and haven't been yet demonstrated in a fully mode-unified environment.