Control of a robotic prosthesis simulation by a closed-loop intracortical brain-machine interface

Abstract

Closed-loop brain-machine interfaces may help restore the autonomy of amputees and tetraplegic patients. However, additional efforts are needed towards their realworld use with prostheses. Here we have interfaced a highly versatile closed-loop mouse BMI with an online model of a realworld prosthetic arm. We describe this setup and illustrate how it allows to explore the efficiency of different input and output coding strategies given a realistic modelling of the interactions between a commercial bidirectional prosthesis and its environment.

Continuity within the somatosensory cortical map facilitates learning

“…Stimulation patterns consisted of light bars 700 microns long and 150 microns wide, rotating on a disk of diameter 1.5 mm. This pattern was chosen as part of our interest in encoding the angle of a joint (Goueytes et al, 2019). The disk was centered on the C2 barrel location (vS1, Figure 1B, n = 11 mice).…”

“…The bar angular position was updated every 10 ms. Each trajectory was taken from a database of 252 pre-loaded trajectories. These were obtained from a closedloop BMI study performed with different mice (Goueytes et al, 2019). In that study, the activity of motor cortex neurons was used to drive the rotating bar.…”

Learning in a closed-loop brain-machine interface with distributed optogenetic cortical feedback

“…Thanks to this software bridge, the speed and direction commands for the proximal joint of the virtual prosthesis was updated approximately every 12 ms, based on the neuronal activity readout. In return, the current angular position of the joint was fed back to the brain-machine interface and was used to update the angular position of the optogenetic S1 stimulation (Goueytes et al, 2019).…”

“…Strategies to provide behaviorally-relevant input using such cortical stimulation often rely on the intensity or frequency modulation of a stimulation targeting one spatially limited region of interest, which limits the amount of information that can be delivered (O'Doherty et al, 2011;Prsa et al, 2017). However, recent technical progress has made distributed neuronal activations possible, by using sophisticated multichannel electrical microstimulations (Dadarlat et al, 2015;Fernández et al, 2021;Flesher et al, 2021;Weiss et al, 2019) or by harnessing spatiotemporally patterned optogenetic stimulation of the cortex (Abbasi et al, 2018;Ceballo et al, 2019;Goueytes et al, 2019;Lassagne et al, 2022). Such distributed neuronal activation at the surface of the cortex can convey multiple information streams in parallel (Hartmann et al, 2016), such as those arising from the multiple touch-like sensors that are available in modern bidirectional prostheses (D'Anna et al, 2019).…”

“…Here, based on a recently developed closed-loop brain-machine interface (Goueytes et al, 2019) we asked if this feedback could be efficiently exploited by the mice beyond sensory processing, by helping to control the angular position of a simulated prosthesis. We found that the mice were able not only to detect the location of the Rewarded zone, but also to alter the dynamics of the rotary joint.…”