Machine Learning techniques in soft robotic suits for whole body assistance

摘要

Soft robotic suits, or exosuits, are progressively spreading in multiple directions with special emphasis for what concerns human assistance and functional rehabilitation: lightweight, portability, and ergonomics are the main advantages with respect to the rigid exoskeletons and play a major role in the growing interest from the scientific community for soft wearable technology. Yet, governing wearable robotics still poses several challenges: controlling soft devices in a robust manner implies to opportunely detect wearer’s movement intention and consequently generating the right amount of assistance. The inclusion of machine learning (ML) algorithms as upper layer in the control loop to enhance robustness and reliability is a feasible solution still not fully explored. Here, we present a soft robotic suit to assist both upper and lower limbs i.e. hand grasping and locomotion through machine learning frameworks running in parallel to a classic control architecture. We tested the system on healthy participants performing functional manipulation tasks and overgound walking for upper and lower limbs respectively. Results demonstrated that the use of machine learning provides a more robust prediction of the interaction between the wearer and the devices than in conditions when only classic control is employed. Such effects are detected as significant reductions of muscular activities as well as lower metabolic rates.