This repository contains the code to implement a novel rehabilitation approach for motor impairments after stroke, inspired by previous work by Sadtler et al., 2014. Patients are initially assessed by measuring finger enslaving patterns during isometric finger presses, similar to Xu et al., 2023. This allows to establish the residual control space that the patient is able to explore (i.e., the intrinsic force manifold). During the subsequent rehabilitation, patients are asked to control a screen cursor through multifinger isometric force patterns. The force patterns generated by the patients are mapped onto the cursor velocity through a linear decoder. Initially, the decoder requires that the force patterns remains within the intrinsic manifold. Subsequently, the mapping is gradually rotated to encourage the patient to explore the portion of the control space outside the intrinsic manifold by generating force patterns outside their initial repertoire. In this way, the approach maximises the chance of engaging the latent corticospinal projections that control force patterns that the patient cannot yet achieve.

On the occasion of Brain Hack 2026 at Western, we presented a series of simulation experiments to evaluate this approach. First, we generated basis vectors simulating the motor repertoire in stroke patients and control participants, and replicated recent results by Xu et al., 2023 (see baseline). From the force patterns generated during simulated single-finger movements, we determined the residual control space that patients were able to span, i.e., the intrinsic force manifold. Next, we trained a feedback controller to move a 2D cursor and reach radially arranged targets using the basis vectors corresponding to stroke patients and control participants. The multi-finger force patterns generated by the controller were mapped onto the velocity of the cursors thorugh a linear decoder that could operate within or outside the intrinsic manifold (see training). Finally, we repeated the training procedure but this time we updated trial-by-trial the basis vectors with reinforcement learning to simulate plastic changes in the motor system throughout rehabilitation. This procedure resulted in the overall improvement of the motor repertoire of stroke patients (see post_rehab).