¹Enchanted Tools, Paris, France
²CIAMS Université d'Orléans, Orléans, France
³CEA List, Grenoble, France
⁴CIAMS, Université Paris-Saclay, Bures-sur-Yvette, France
⁵LURPA, ENS Paris-Saclay, Université Paris-Saclay, Gif-sur-Yvette, France
⁶Université Rennes 2, Rennes, France
⁷Imperial College London, London, United Kingdom

Text

Active exoskeletons have been employed in neurorehabilitation and to assist workers with physically demanding tasks. However, this approach does not necessarily promote more efficient or ergonomic postures, and may promote harmful postures, as the exoskeleton compensates for the physiological costs. The present study aims at designing a method leveraging active exoskeletons to guide users in adopting and retaining ergonomic postures. Our method is thought as an alternative to traditional biofeedback by simultaneously (i) measuring posture, and (ii) generating sensory stimulation through interaction landscapes. Thus, the exoskeleton performs both the sensing and sensory stimulation components of biofeedback, removing the need for additional devices. Importantly, the landscapes should not be a simple trajectory guidance, as there is evidence that humans learn more efficiently when they can freely explore the task space and when motor errors occur.

We designed a reaching experiment where: (i) the target was a long vertical rectangle, allowing exploration, (ii) movements were constrained in a parasagittal plane, (iii) the landscape consisted of horizontal vibrations pushing participants outside of the target, and (iv) their amplitude was null either 20 cm above or below the preferred reaching height of each participant. These two groups were formed depending on whether this preferred height was near, in which case the minimum was below preferred height (below group), or below, with minimum above preferred height (above group) shoulder height. For the above group, minimizing vibrations increased gravity-related efforts, and conversely for below. Participants first performed 2 blocks with the exoskeleton in transparent mode, followed by 4 blocks with the vibrations, and 2 blocks to analyze after-effects.

Both groups gradually learned to minimize the vibrations, until they were sufficiently small to remain within the target at no additional effort cost. This supports our assumption that humans can explore vibration landscapes, thereby performing a “gradient descent” to minimize disturbances. The below group rapidly decreased gravity-related efforts in reaction to the vibrations, thereby improving their posture. More surprisingly, the above group increased gravity-related efforts to reduce vibrations, showing the priority assigned to successful task completion by participants at the cost of less ergonomic postures. Finally, both groups exhibited lower gravity-related efforts during the after-effects blocks, with a clear retention for the below group, suggesting that exploring a variety of postures, whether better or worse than theirs, allows users to minimize the efforts from a task.

Our approach solves several issues of traditional biofeedback at the cost of using an active exoskeleton. On the sensing side, it does not require any other sensors than those embedded in the exoskeleton, which drastically limits the installation and calibration complexity. On the sensory side, we only stimulate the sensory streams already involved in performing the task, minimizing the additional cognitive load.

Finally, this stimulation can easily be transposed to any task or posture, without increasing the complexity of the setup.