A biomechanical model for encoding joint dynamics: applications to transfemoral prosthesis control
This paper presents and tests a framework for encoding joint dynamics into energy states using kinematic and kinetic knee joint sensor data and demonstrates how to use this information to predict the future energy state (torque and velocity requirements) of the joint without a priori knowledge of the activity sequence. The intended application is for enhancing micro-controlled prosthetics by making use of the embedded sensory potential of artificial limbs and classical mechanical principles of a prosthetic joint to report instantaneous energy state and most probable next energy state. When applied to the knee during preferred and fast speed walking in 8 human subjects (66 preferred-speed trials and 50 fast-speed trials), it was found that joint energy states could be consistently sequenced (75% consensus) according to mechanical energy transference conditions and subsequences appeared to reflect the stability and energy dissipation requirements of the knee during gait. When simple constraints were applied to the energy transfer input conditions (their signs), simulations indicated that it was possible to predict the future energy state with an accuracy of >80% when 2% cycle in advance (∼20 ms) of the switch and >60% for 4% (∼40 ms) in advance. This study justifies future research to explore whether this encoding algorithm can be used to identify submodes of other human activity that are relevant to TFP control, such as chair and stair activities and their transitions from walking, as well as unexpected perturbations.