Co-adaptive learning and domain adaptation bridge the translational gap in neuroprosthetic control: a review

Myoelectric and neural prosthetic devices frequently fail in daily clinical use despite laboratory success, a persistent translational gap driven by domain shift, the non-stationary nature of biological signals arising from neural plasticity, motor learning, muscle fatigue, and electrode displacement. This comprehensive review synthesizes the neurophysiological and technical origins of domain shift and evaluates algorithmic strategies developed to accommodate it. We identify that domain shift stems from coupled biological and technical processes, including cross-session drift and anatomical variability. Algorithmic countermeasures, such as closed-loop decoder adaptation, adaptive filtering, ensemble modeling, reinforcement learning, and transfer learning, address distinct facets of this problem, balancing adaptation speed against system stability. Crucially, we emphasize co-adaptive systems where decoder updates interact directly with ongoing neural and motor plasticity. Evidence indicates that treating the user as a passive signal source limits long-term control; instead, the user must be viewed as an active learning partner. Sustained improvements in neuroprosthetic reliability require integrating multiple adaptation strategies within a translational neuro-engineering framework that respects biological timescales. This integrated approach may help reduce long-term prosthetic device abandonment and advance accessible, robust assistive neurotechnology for clinical populations worldwide.