User-Tailored Learning to Forecast Walking Modes for Exosuits

Abstract

Assistive robotic devices, like soft lower-limb exoskeletons or exosuits, are widely spreading with the promise of helping people in everyday life. To make such systems adaptive to the variety of users wearing them, it is desirable to endow exosuits with advanced perception systems. However, exosuits have little sensory equipment because they need to be light and easy to wear. This paper presents a perception module based on machine learning that aims at estimating 3 walking modes (i.e., ascending or descending stairs and walking on level ground) of users wearing an exosuit. We tackle this perception problem using only inertial data from two sensors. Our approach provides an estimate for both future and past timesteps that supports control and enables a self-labeling procedure for online model adaptation. Indeed, we show that our estimate can label data acquired online and refine the model for new users. A thorough analysis carried out on real-life datasets shows the effectiveness of our user-tailored perception module. Finally, we integrate our system with the exosuit in a closed-loop controller, validating its performance in an online single-subject experiment.

Figures (9)

• Figure 1: We equip an easy-to-wear and light exosuit with a user-tailored perception model estimating the current and intended walking gait (ascending or descending stairs and walking on level ground) using only IMU readings.
• Figure 2: Proposed approach: at each timestep $k$, the model $\bm{m}$ is fed with an input window of the last $M+1$ samples and outputs an estimate of walking modes in a target window including $N$ timesteps ahead and behind; $\delta$ marks the timestep in the target window where $\bm{m}$ performs best. The estimate $\hat{c}_{k+\delta}$ is used to pseudo-label the corresponding input $\bm{f}_{k+\delta}$ and refine $\bm{m}$.
• Figure 3: Exosuit's main components and control architecture.
• Figure 4: A simplified architecture of our tcn model with the kernel size set to $2$ and $N=1$. Two hidden layers output $4$ and $8$ feature channels, respectively. The dotted gray lines represent dilated causal convolutions. Black arrows highlight those producing $\hat{\bm{y}}_{k}$ (rightmost blue circle). This is the target window estimate resulting from the two-channel input window (gray circles on top) containing $\bm{f}$ from time $k$ back to $k-M$.
• Figure 5: Confusion matrix for the rf (left) and tcn approach (right).