User-Tailored Learning to Forecast Walking Modes for Exosuits
Authors: Gabriele Abbate, Enrica Tricomi, Nathalie Gierden, Alessandro Giusti, Lorenzo Masia, Antonio Paolillo
Organizations: Dalle Molle Institute for Artificial Intelligence (IDSIA), USI-SUPSI, Lugano, Switzerland · Munich Institute for Robotics and Machine Intelligence (MIRMI), Department of Computer Engineering, School of Computation, Information and Technology, Technical University of Munich (TUM), Munich, Germany.
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 & tables
Fig. 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.
Fig. 2: Proposed approach: at each timestep k , the model 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; δ marks the timestep in the target window where m performs best. The estimate c^k+δ is used to pseudo-label the corresponding input fk+δ and refine m .
Fig. 3: Exosuit’s main components and control architecture.
Fig. 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 y^k (rightmost blue circle). This is the target window estimate resulting from the two-channel input window (gray circles on top) containing f from time k back to k−M .
Fig. 5: Confusion matrix for the rf (left) and tcn approach (right).
Fig. 6: Models’ performance for each timestep in the target window: auroc of the tcn (solid blue) and tf (solid red); tcn , tf and rf repeating the estimate at k in the future (dashed blue, red and gray dash-dotted lines, respectively). The dotted black line at timestep k+δ corresponds to the peak of performance for the tcn model.
Fig. 7: tcn inference for a 14s time-frame: IMUs signals composing the feature vector, with the input window (top plot, gray lines) and prediction of the model compared against the ground truth (bottom plot, blue and orange lines respectively). In the target window, the probabilities of each walking class are depicted with blue shades (the darker, the higher).
Fig. 8: Models’ performance over time: incrementally adding labeled data to the training set improves the auroc (orange line) even using our self-labeling procedure (blue). Error bars denote 95% confidence intervals.
Fig. 9: Snapshots of an experiment running our perception model integrated in closed-loop with the exosuit’s controller. The top plot shows the recorded ground truths for the walking mode (orange) and the model’s classifications (blue). The call-out rectangle at the bottom shows in detail the first part of the experiment, in particular: input features signals (top plot), walking mode ground truth and estimates (middle), and reference motor positions (bottom). The purple and pink lines show that the amplitudes of the motor references are correctly modulated according to the current walking mode.
University of California, Los Angeles Los Angeles, California, USA · Columbia University New York, New York, USA · University of Wisconsin-Madison Madison, Wisconsin, USA +1