Direct neural interface for controlling the movement of an effector

Abstract

A method for controlling the displacement of an effector in space by means of a direct neural interface, comprising: a) acquisition of electrophysiological signals produced in the cortex of an individual at a measurement time; b) processing of the electrophysiological signals to form input data; c) processing of the input data by a predictive model implemented by the neural interface to define a displacement of the effector at the measurement time; d) determination of a corrected displacement of the effector from the displacement defined by the predictive model in step c); e) displacement of the effector in space by the direct neural interface, based on the corrected displacement resulting from step d; f) repetition of steps a) to e), the method comprising, in each step d), an estimation of the target position at the measurement time. Figure 2.

Description

Title of the invention: Direct neural interface for controlling the movement of an effector Technical field

[0001] The invention relates to a direct neural interface, usually designated by the English acronym BCI, meaning Brain Computer Interface, to enable the control of an automated effector, for example a robot or an exoskeleton, by cortical electrophysiological signals. PREVIOUS ART

[0002] The field of direct neural interfaces is in full development and appears to be an attractive solution to allow disabled people to control effectors by thought. This involves recording, generally using cortical electrodes, electrophysiological signals. These are processed by algorithms allowing a control signal to be extracted, to control effectors. The control signal can allow the piloting of an effector of the exoskeleton, computer orthosis or robot type, to provide assistance to the user. The algorithms implemented allow a translation of an instruction given by the user, this instruction being captured, by the electrodes, in the form of so-called electrophysiological signals, because they are representative of the electrical activity of the neurons. This electrical activity is generally measured at the level of the cortex, by means of cortical electrodes arranged in the cranium.It can also be measured by electroencephalography electrodes, which are less intrusive because they are placed on the scalp, but also less efficient, particularly in terms of spatial resolution. Another solution is to record electrophysiological signals by magnetoencephalography, which requires a dedicated installation.

[0003] The algorithms implemented are generally based on a predictive model. The predictive model uses input data, obtained by preprocessing the recorded electrophysiological signals. An example of calibration of a predictive model is described in US11497429 (this is DD17618). The predictive model allows processing of the input data to establish a predicted displacement, making it possible to perform a displacement of the effector corresponding to the detected electrophysiological signals.

[0004] However, the use of a predictive model can be accompanied by a certain inaccuracy, when the user seeks to move the effector to a predetermined target position. The invention makes it possible to solve this problem. Statement of the invention

[0005] A first object of the invention is a method for controlling the movement, in a space, of an effector, by a direct neural interface, the movement tending to bring the effector closer to a target position, comprising the following steps: - a) acquisition of electrophysiological signals produced in the cortex of an individual, at a measurement time, the effector occupying a position, in space, at said measurement time; - b) processing of electrophysiological signals, to form input data; - c) processing of the input data, by a predictive model implemented by the neural interface, to define a displacement of the effector at the measurement instant; - d) determination of a corrected displacement of the effector, from the displacement defined by the predictive model during step c); - e) controlling the movement of the effector in space, by the direct neural interface, from the corrected movement resulting from step d); - f) repeating steps a) to e), by incrementing the measurement time, until a criterion for stopping the iterations is reached;

[0006] the method comprising, during each step d) following at least a first iteration of steps a) to e): - di) estimation of the target position, at the measurement time, from at least: • a position of the effector at the measurement time, as well as at least one time preceding the measurement time; • a displacement defined by the predictive model at the measurement time as well as at least the time preceding the measurement time; - due) from the estimated target position during sub-step di), determination of a movement towards the estimated target; - diii) taking into account the displacement towards the estimated target, resulting from sub-step dii), to establish the corrected displacement at the measurement time.

[0007] Sub-step di) may comprise: - calculation of a target probability density, corresponding to a probability density that each point in space corresponds to the target position; - determination of the point in space maximizing the target probability density, or a quantity comprising the target probability density, the point thus determined corresponding to the estimated target position.

[0008] The calculation of the target probability density can take into account: • several positions of the effector at times preceding the measurement time; • several displacements defined by the predictive model at times preceding the measurement time.

[0009] The calculation of the target probability density can be established from a combination, for example linear, - of a first conditional probability density conditioned by • positions of the effector at times preceding the measurement time; • movements defined by the predictive model at times preceding the measurement time; - and a second conditional probability density conditioned by: • the position of the effector at the time of measurement; • the displacement defined by the predictive model at the measurement time.

[0010] The first conditional probability density can be weighted by a forgetting factor.

[0011] Sub-step di) may comprise a determination of the point maximizing a quantity comprising the target probability density, the quantity corresponding to the reduced target probability density: • a first distance, weighted by a first regularization factor, between each point and the target position estimated at a previous measurement time; • and / or a second distance, weighted by a second regularization factor between each point and the position of the effector at the measurement time.

[0012] During sub-step dii), the movement towards the estimated target can be established so that the interface moves the effector from the position at the measurement time to the target position estimated at the measurement time.

[0013] In sub-step diii), the corrected displacement may be a combination, for example a linear combination, of the displacement defined by the predictive model in step b) and the displacement towards the estimated target resulting from sub-step dii). The linear combination may implement: - a first weighting factor, applied to the displacement defined by the predictive model; - a second weighting factor, applied to the movement towards the estimated target.

[0014] A second object of the invention is a direct neural interface, comprising: - sensors, capable of detecting electrophysiological signals representative of cortical activity; - an effector, configured to be actuated by a control signal generated by the direct neural interface; - a processor programmed to determine a displacement of the effector based on input data resulting from the electrophysiological signals detected;

[0015] the processor being configured to implement steps c) to f) of a method according to the first subject of the invention, at different measurement times.

[0016] The invention will be better understood upon reading the description of the exemplary embodiments presented in the remainder of the description, in conjunction with the figures listed below. FIGURES

[0017] [Fig.l] schematizes the main components of a direct neural interface.

[0018] [Fig.2] represents a displacement of an effector in a 3D environment.

[0019] [Fig.3] illustrates a vector corresponding to a displacement predicted by a predictive model, a desired displacement, between the position of the effector and the position of the target, as well as the angular error between the two vectors.

[0020] [Fig.4] shows the main steps of a method implemented by a neural interface according to the invention.

[0021] Figures 5A and 5B show the evolution of distances between the effector and the target, as well as between the estimated target and the actual target, as a function of time, considering two different noise models. In Figures 5A and 5B, the effector is moved towards the target without taking into account the estimated positions of the target over time.

[0022] Figures 6A and 6B show the evolution of angular precisions as a function of time, considering two different noise models.

[0023] Figures 7A to 7D show the influence of the weighting of the displacement of the effector between the displacement predicted by the interface and a corrected displacement, by implementing the invention, as a function of an estimation of the position of the target.

[0024] [Fig.8] illustrates an implementation of the invention in 3D. PRESENTATION OF SPECIAL EMBODIMENTS

[0025] [Fig.l] represents the main elements of a direct neural interface 1 according to the invention. It is a device comprising sensors 2, allowing the acquisition of electrophysiological signals representative of a neuronal activity. The sensors 2 are for example cortical electrodes Ei... .En. The sensors 2 are connected to a processor 3, by a wired or wireless link. The processor 3 is able to execute instructions coded in a memory 4. These instructions include in particular the algorithms allowing the generation of a control signal intended to control an effector 5.

[0026] The effector 5 is an actuator capable of performing an action under the effect of the control signal addressed to it. It may be an exoskeleton, a robot, or a computer. The effector is then controlled by the electrophysiological signals acquired by the cortical electrodes 2, via an algorithm coded in the memory 4 and implemented by the processor 3. More precisely, the objective of the device is to allow a movement of the effector, in an environment, to a target position.

[0027] The environment may be a real environment, the effector being for example a robotic arm, a robotic orthosis, a glove or a tool carried by the user's hand. The objective is to move the effector to the target position, which is a position that the user wishes to reach. The environment may be virtual, the effector being for example a computer mouse. The user can then control a cursor to a determined position on a screen, so as to activate a command. The target position is the position of the cursor that the user wishes to reach.

[0028] The device 1 is a direct neural interface, in the sense that it allows the control of the effector 5 by the electrophysiological signals measured by the sensors 2. The electrophysiological signals can be acquired by cortical electrodes placed in contact with the sensorimotor cortex, magnetoencephalography sensors or electroencephalography electrodes.

[0029] The control signal is formed by the processor 3 from the electrophysiological signals. The algorithm implements a predictive model, generating an output signal, allowing the control of the effector 5. The predictive model can be determined as described in US patent 11497429. More precisely, the output signal makes it possible to move the effector 5 to the target position according to a displacement predicted by the predictive model.

[0030] The space can be discretized into a plurality of points. Figure 2 illustrates a progressive displacement of an effector between an initial position and a target position Xpo. The superscript tr designates the test carried out. The displacement is carried out jerkily, depending on the sampling frequency of the interface, the latter possibly being 10 Hz. The sampling frequency is preferably fast enough not to appear too jerky for the user, and to limit the reaction time of the system when the user changes intention. At each measurement instant 5, the effector occupies a point position x£. x£. is a point represented by a vector whose dimension corresponds to the dimension of the space in which the effector evolves. In a first embodiment, a 2-dimensional space is considered. Each vector is defined by an x ​​coordinate and an X coordinate

[0031] At each instant, the predictive model of the interface generates a predicted displacement Xp, which corresponds to the displacement corresponding to the electrophysiological signals detected by the interface, ^p = x^1 - x£ (1). The displacement Xp corresponds to the displacement of the effector, as predicted by the predictive model, between two successive instants s, s+l, as a function of the electrophysiological signals detected at instant J.

[0032] At each instant, we can determine a vector which corresponds to a desired displacement, such as:

[0033] xj= (2)

[0034] However, the position , although known by the user, is not known by the interface. It is therefore necessary to estimate it, which is described below, so as to be able to determine 3¾. x^ corresponds to an ideal displacement vector, which would be followed by the effector without the imperfections of the interface. We assume that the user intends to move the effector towards the target in a straight line.

[0035] In Figure 3, the vectors x^ and Xp are represented. The angle between the two vectors x^ and Xp corresponds to an angular error, noted a(xp, Xj)-

[0036] An important element of the invention is the estimation, by the interface, at each measurement instant, of the position of the target. The estimation of the position of the target is obtained by maximizing a probability, called target probability, defined at any point in space, and corresponding to the probability that each point in space corresponds to the target x|Ç .

[0037] The target probability is a conditional probability, defined at several points in space, and preferably at any point in space. Assuming that the user wishes to command a movement of the effector in a straight line towards the target position, and therefore seeks to reduce the angular error, it can be estimated that for each point in space, represented by a vector x, with coordinates (x' 3); the angular error a(Xp, X - x£) is a variable following a Gaussian probability density, with variance o2.

[0038] P(x = X;C | x;', ) = -=J= e-®

[0039] The target probability density P(x = x[f | x^, Xp ) is estimated at each measurement time '.

[0040] The variance o2 can be estimated as described below.

[0041] Advantageously, the target probability density can be estimated by taking into account the positions of the effector x^1... x^ and the displacements predicted by the predictive model Xp"1... Xp at times preceding the measurement time. Thus, the target probability density is such that:

[0042] p(x = xtr |xi Xp ) = ~Ei=iP(x= x^|xL xf, )

[0043] For a point of position x, the probability density according to expression (4) corresponds to an averaging of the target probabilities established over time for said point.

[0044] At each instant, the position of the target is considered as that maximizing the target probability density.

[0045] = arg max[P(x= jx^... x*., x|... x), ) ] (5)

[0046] At each instant the estimation of the position of the target is used to control the movement of the effector. A movement towards the estimated target is established, such that: [°047] <-x^(6):

[0048] X'p mpt is the vector between the current position x£ and the estimated position of the target. This is the displacement vector towards the estimated target.

[0049] is the vector between the current position xg and the actual position of the target x^. H is the desired displacement vector.

[0050] The position of the effector at time x£+1 can be estimated by the expression:

[0051] ! -ah + 2^L.. \ (7) c - / T IM U} 114^11 r Xf?

[0052] Where k is a first weighting factor and 1 - k is a second weighting factor. In this example, 0 <k<l. fl est un réel positif, correspondant à une distance de déplacement préalablement déterminée. Choisir k = 1 revient à ne pas prendre en compte la position estimée de la cible, ce qui revient à un déplacement selon l’art antérieur. Choisir k = 0 revient à ne prendre en compte que l’estimation de la cible dans la commande du déplacement.

[0053] The first weighting factor k and the second weighting factor 1 - k make it possible to measure the contributions: - the displacement predicted by the predictive model of the interface; - movement towards the estimated position of the target.

[0054] The influence of these weighting factors is discussed in connection with Figures 7A to 7D.

[0055] The displacement x^1 - xg corresponds to a corrected displacement, taking into account both the displacement predicted by the predictive model, and the displacement towards the estimated position of the target. This is a key aspect of the invention.

[0056] Thus, the movement of the effector is partly defined by the predictive model, as in the prior art, and partly defined as a function of the estimated position of the target. Taking into account the estimated position of the target, in determining the movement of the effector, is an important aspect of the invention.

[0057] According to a variant, the target probability density is such that:

[0058] (8)

[0059] where k is a positive real number, corresponding to a forgetting factor, k makes it possible to dose the taking into account of previous instants in relation to the current instant s. A is generally between 0 and 1. makes it possible to modulate the impact of old instants in taking decision.

[0060] According to this variant, the target probability density is established from a linear combination: • of a first conditional probability density P( x — xta |x(l... xf'1 , xj>... xJ;1 ), weighted by the forgetting factor and conditioned by the positions of the effector at times preceding the measurement time, as well as by displacements predicted by the predictive model at times preceding the measurement time. • a second conditional probability density P(x = x^ | xf, , conditioned by the position x£ of the effector at the measurement time and the displacement predicted by the predictive model Xp at the measurement time.

[0061] According to another variant, the estimation of the position of the target, at each measurement instant, is carried out according to the expression:

[0062] = P(x= xtg|xi... x^,x^... x^ ) - || xf^-x || -d2 || x*-x || ] ^9)

[0063] Where and ^2 are positive scalars, which correspond to regularization factors.

[0064] The aim here is to maximize a quantity established from the target probability density, reduced: - a first distance, between the position of the target ^Ù and each point x in space, said first distance being weighted by a first regularization factor. This makes it possible to avoid two successive estimated positions of the target being too far from each other; - and / or a second distance between the position of the effector X* at the measurement time and each pointx in space, said second distance being weighted by a second regularization factor 22. This makes it possible to prevent the estimated position of the target from being too far from the position of the effector at the measurement time. This makes it possible to contain the position of the target in a certain working space, around the current position.

[0065] Expression (9) can be implemented without taking into account the first distance ( / ^ = 0) or without taking into account the second distance (¾ = 0).

[0066] [Fig.4] shows schematically the main steps of a method implemented by a direct neural interface according to the invention.

[0067] Step 100: taking into account electrophysiological signals, emitted by the user and detected by the sensors 2 at a time s;

[0068] Step 110: from the detected electrophysiological signals, formation of an input signal feeding the predictive model.

[0069] Step 120: definition, by the predictive model, of the displacement Xp at the instants, in function of the input signal.

[0070] Step 130: determination of a target probability density, P(x = x^|x£... x^,x|... x^> ), P(x= x^|xS, x| ) at any pointx in the space in which the effector is moving, as a function of the position of the effector at time s, of the predicted displacement at time s, and possibly as a function of the position of the effector and of the predicted displacement at times preceding time s: cf. expressions (3), (4) or (8).

[0071] Step 140: estimation of a position maximizing the probability density resulting from step 130, or of a quantity using the probability density resulting from step 130. See expressions (5), (6) or (9).

[0072] Step 150: from the calculation of a displacement towards the estimated position of the target Xpjy[pT

[0073] Step 160: from the displacement Xp defined by the predictive model, resulting from step 120, and from the displacement towards the estimated position of the target, resulting from step 150, determination of a corrected displacement x^1 - x£: cf. expression (7)

[0074] Step 170: displacement of the effector according to the corrected displacement.

[0075] Step 180: incrementing the time index and repeating steps 100 to 170 until the target is reached or is considered to be reached, or until the algorithm is voluntarily stopped by the user. Determination of o

[0076] The standard deviation o is previously determined during preliminary tests. During these tests, the position x^Ç of the target is known. We can establish a cost function allowing us to estimate an angular precision, and determine the value of o which optimizes the cost function. The cost function can be an average cosine similarity function MCS such that: 100771 MCS = ^^'L'^CosSim^ (11)

[0078] where - Nss corresponds to a number of preliminary tests carried out; - st>- corresponding to the number of time increments during each test; - CosSim^Xp, x^) corresponds to the cosine similarity function, such that: CosSim^p, x^ = (U} ; = xL - x* ; - Xj = Xp, - xsc. Xj corresponds to the desired displacement.

[0079] The value of o is the one that maximizes the MCS cost function.

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095] The cost function can be an average error of the estimated target position relative to the actual target position, over the last time steps, for example the last 30 time increments. = -29 II “x'9 II and having been defined in connection with (11). The value of o is the one that minimizes the MDE cost function. Simulations Simulations were carried out on artificial data sets, in which the position of the target was known. Thus, at each instant, is known. From Xj, we simulated the displacement predicted by the predictive model of an interface by considering that the displacement predicted by the predictive model Xp is such that: Xp = (13) where is a vector representing noise. Two different noise models were considered: a first noise model was an independent and identically distributed (IID) Gaussian noise of power q2=1.4. A second noise model was a Gaussian noise dependent on the previous time increments, such that: A = E10,0.951&-^vv'-' (14) Where w follows a centered IID law of power q2=0.6. The target probability density was calculated by considering o2 = 3 rad2. A first series of tests was carried out considering k = 1 in (7). During these tests, a target position was estimated at each time increment. The estimated target position was not used in the control of the effector displacement. [Fig.5A] shows the evolution, as a function of time, of the distance between the effector and the target (curve a) as well as the distance between the estimated target and the real target (curve b), considering the first noise model. The abscissa axis corresponds to time and the ordinate axis corresponds to the studied distances. [Fig.5B] shows the evolution, as a function of time, of the distance between the effector and the target (curve a) as well as the distance between the estimated target and the real target (curve b), considering the second noise model. The abscissa axis corresponds to time and the ordinate axis corresponds to the studied distances. In Figures 5A and 5B, the dashed curves represent the limits of the confidence intervals. In both figures, we observe that the position of the target is stabilized when the effector is sufficiently close to the target. '>60 in [Fig.5A] and s>40 in [Fig.5B]

[0096] Figure 6A shows the evolution, as a function of time, of the angular precision calculated by taking into account the predictive model (curve a) and the estimation of the target (curve b), by considering the first noise model. The abscissa axis corresponds to time and the ordinate axis corresponds to the angular precision, expressed in the form of a cosine similarity. Curve a corresponds to CosSimi^, Xp ) and curve b corresponds to CosSim^X^. i^PT ) • This curve makes it possible to compare the prediction performance: - of the predictive model alone (curve a), which amounts to implementing (7) with k = 0; - the estimation of the position of the target, which amounts to implementing (7) with k = 1.

[0097] Figure 6B shows the evolution, as a function of time, of the angular precision calculated by taking into account the predictive model (curve a) and the estimation of the target (curve b), by considering the second noise model. The abscissa axis corresponds to time and the ordinate axis corresponds to the angular precision, expressed in the form of a cosine similarity. As in Figure 6A, curve a corresponds to CosSim(x^, Xp ) and curve b corresponds to CosSim(x^ x^pt )•

[0098] It is observed that the angular precision obtained by the estimation of the target by the method proposed here (curves b) is better compared to that resulting from the predictive model, and this as soon as the effector is sufficiently close to the target: s>60 on [Fig.6A] (cf. [Fig.5A]) and s>40 on [Fig.6B] (cf. [Fig.5B]).

[0099] Figures 7A to 7D represent the evolution, as a function of time, of the distance between the effector and the real target (curve a) as well as the distance between the estimated target and the real target (curve b), considering the second noise model. The abscissa axis corresponds to time and the ordinate axis corresponds to the studied distances. The value of k was varied in the different figures: [Fig.7A]: k=0.9; [Fig.7B]: k=0.7; [Fig.7C]: k=0.5; [Fig.7D]: k=0.

[0100] Reducing k makes it possible to better stabilize the position of the effector near the target, at the cost of a certain delay in convergence.

[0101] The curves of Figures 7A to 7D can be compared to Figure 5B, the latter corresponding to a displacement of the effector without taking into account the estimated position of the target (k = 1). Three-dimensional embodiment

[0102] The invention has been described by considering a movement of the effector in two dimensions. The same principles can be applied in three dimensions. Figure 8

[0103]

[0104]

[0105]

[0106]

[0107]

[0108]

[0109]

[0110] [YES] shows the angles to be taken into account, in a three-dimensional frame of reference, according to which each position x is defined by three coordinates (x, yz) (p = arctan(^ ) (15) 0 = arctan — 16 The angular errors between two vectors xi, with coordinates y 7 J and x2 with coordinates (^ y are such that: ar(Xj, x2) = arctan^ ) -arctan^ ) And ✓ \ . / ^'r+y2 \ \ (18) ae(x1? x2) = arctan ......- arctan ....... Assuming that the user seeks to reduce angular errors at each time step, and that the angular errors follow independent Gaussian distributions of the same power, the target probability density, defined in 2D in expression (3), becomes: P(x = x^xf.xj, ) = 2^7 (19) Steps 100 to 180, previously described, can be implemented in 3 dimensions from the target probability density defined in (19).

Claims

1. Claims Method for controlling the movement, in a space, of an effector by a direct neural interface, the movement tending to bring the effector closer to a target position, the method comprising the following steps: - a) acquisition of electrophysiological signals produced in the cortex of an individual, at a measurement time, the effector occupying a position, in space, at said measurement time; - b) processing of electrophysiological signals, to form input data; - c) processing of the input data, by a predictive model implemented by the neural interface, to define a displacement of the effector (Xp) at the measurement time; - d) determination of a corrected displacement (xg+1 - xJ) of the effector, from the displacement defined by the predictive model (Xp) during step c); - e) controlling the movement of the effector in space, by the direct neural interface, from the corrected movement resulting from step d); - f) repeating steps a) to e), by incrementing the measurement time (s), until a criterion for stopping the iterations is reached; the method comprising, during each step d) following at least a first iteration of steps a) to e),: - di) estimation of the target position (x^), at the measurement time, from at least: • a position of the effector (x£) at the measurement time, as well as at least one time preceding the measurement time; • a displacement defined by the predictive model (x^) at the measurement time as well as at least the time preceding the measurement time; - due) from the estimated target position during sub-step di), determination of a movement towards the estimated target ( - diii) taking into account the displacement towards the estimated target (resulting from sub-step dii), to establish the corrected displacement (x^1 - x£) at the measurement time.

2. Method according to claim 1, in which sub-step di) comprises: - calculating a target probability density, corresponding to a probability density that each point in space corresponds to the target position; - determining the point in space maximizing the target probability density, or a quantity comprising the target probability density, the point thus determined corresponding to the estimated target position.

3. Method according to any one of the preceding claims, in which the calculation of the target probability density takes into account: • several positions of the effector (x^... x£) at times preceding the measurement time; • several displacements defined by the predictive model (Xp... Xp) at times preceding the measurement time.

4. Method according to claim 3, in which the calculation of the target probability density is established from a combination: - of a first conditional probability density ( P(x = xfg |x|... xg4 , xj>... x^1 ) ), conditioned by • positions of the effector at times preceding the measurement time; • displacements defined by the predictive model at times preceding the measurement time; - and of a second conditional probability density, P( x = xîg [ x£, x), ) conditioned by: • the position of the effector at the measurement time; • the displacement defined by the predictive model at the measurement time.

5. The method of claim 4, wherein the first conditional probability density is weighted by a forgetting factor ( / *).

6. Method according to any one of claims 2 to 5, in which sub-step di) comprises a determination of the point maximizing a quantity comprising the target probability density, the quantity corresponding to the target probability density reduced by: • a first distance (y _ x ||), weighted by a first regularization factor (20, between each point and the target position estimated at a previous measurement instant; • and / or a second distance ( || x / . - X ||), weighted by a second regularization factor (22), between each point and the position of the effector at the measurement instant.

7. Method according to any one of the preceding claims, wherein during sub-step dii), the movement towards the estimated target is established so that the interface moves the effector from the position at the measurement instant to the target position estimated at the measurement instant.

8. Method according to any one of the preceding claims, wherein in sub-step diii), the corrected displacement is a combination of the displacement defined by the predictive model in step b) and the displacement towards the estimated target resulting from sub-step dii).

9. Method according to claim 8, in which the linear combination implements - a first weighting factor (k), applied to the displacement defined by the predictive model; - a second weighting factor (1-^), applied to the displacement towards the estimated target.

10. Direct neural interface (1), comprising: - sensors (2), capable of acquiring electrophysiological signals representative of cortical activity; - an effector (5), configured to be actuated by a control signal generated by the direct neural interface; - a processor programmed to determine a movement of the effector as a function of input data resulting from the electrophysiological signals detected; the processor (3) being configured to implement steps c) to f) of a method which is the subject of any one of the preceding claims, at different measurement times.

Images