EP2672931A1

EP2672931A1 - Method for determining an artificial periodic patterned signal

Info

Publication number: EP2672931A1
Application number: EP11704756.3A
Authority: EP
Inventors: Matthieu Duvinage; Thierry Castermans; Thierry Dutoit
Original assignee: Universite de Mons Hainaut
Current assignee: Universite de Mons Hainaut
Priority date: 2011-02-10
Filing date: 2011-02-10
Publication date: 2013-12-18
Also published as: WO2012107096A1

Abstract

The present invention is related a method for determining an artificial periodic patterned signal comprising the steps of : providing a set of commands for driving a central pattern generator; - providing a brain computer interface (BCI); determining a set of signals originating from said BCI, said set of signals corresponding to said set of commands; determining from a BCI measurement a signal pertaining to said set of signals; sending a command corresponding to the determined signal to the central pattern generator, said central pattern generator producing an artificial periodic patterned signal based on said command.

Description

METHOD FOR DETERMINING AN ARTIFICIAL PERIODIC PATTERNED SIGNAL Field of the Invention

[0001] The present invention is related to a method for determining an artificial periodic patterned signal.

[0002] The present invention is also related to a system implementing such a method.

[0003] Another aspect of the invention is related to

Lower limb prosthesis or orthosis comprising such a system.

State of the Art

[0004] Current active leg prostheses do not integrate the most recent advances in Brain-Computer

Interfaces (BCI) and bipedal robotics. Moreover, their actuators are seldom driven by the subject's intention.

[0005] Over the years, different kinds of leg prostheses have been developed in order to replace the limb that amputees have lost. The main objective of these prostheses is to allow their user to walk as naturally as possible .

[0006] In fact, the complexity of human walk is such that most of the leg prostheses available on the market today use passive mechanisms. Although these systems are functional, their performance is really limited compared to a real human leg as they do not have self-propulsion capability . [0007] Unfortunately, amputees using this standard technology have to compensate for these limitations. Consequently, they generally develop various strategies which generate reduced locomotion speed, a non-natural gait, considerable fatigue and possibly harmful consequences like recurrent pain and injuries at the interface between their residual limb and the prosthesis.

[0008] Active prostheses solve these problems partially: powered by a battery-operated motor, they move on their own and therefore reduce the fatigue of the amputees while improving their posture. Two main categories of active prostheses exist to date: firstly, devices controlled according to the motion of other healthy parts of the body and secondly, devices equipped with a myoelectric control system.

[0009] In the first category, WO 2004/017873,

EP1848380 and EP1786370 describe sensors placed on the healthy leg of the amputee. By analyzing the motion of the leg with a sophisticated algorithm, the control system can identify the phase of the gait cycle and trigger an actuator to appropriately adjust one or more prosthetic or orthotic joints.

[0010] Instead of exploiting the motion of the healthy leg of the amputee, EP1260201 describes other systems analyzing upper-body motions to trigger and maintain walking patterns.

[0011] The second type of active prostheses (or orthoses) is controlled by myoelectric signals recorded at the surface of the skin, just above the muscles. These signals are then used to guide the movement of the artificial limb. This type of prosthesis is described for example by Y. Sankai, in article "Leading edge of cybernics: Robot suit hal," published in SICE-ICASE, 2006. International Joint Conference, 2006. [0012] The improvement brought by the active prosthetic technology with respect to conventional prostheses is indisputable.

[0013] However, several aspects still need to be improved. For instance, an intuitive interface from which user's intent can be determined is still missing. Additionally, no sensory feedback is provided to the user.

[0014] Active research is being carried out in these two latter areas, in particular for arm and hand prostheses. Complex nerve surgery techniques are being developed as well as new signal processing algorithms and new electrodes, in order to connect an amputee to an artificial limb that he can control intuitively with his own residual nerves and muscles, as described by D. Naidu et Al . in "Control strategies for smart prosthetic hand technology: An overview," in EMBC 2008, Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2008.

[0015] There is therefore a trend in current research to develop prosthesis for amputees wherein the user will have the opportunity to fully recover human mobility and perception, but paying the price of an important and risky surgery. For that reason, more simple systems taking into account the user' s intent are desirable in the meantime.

[0016] Recent researches in the field of Brain-

Computer Interfaces (BCI) have considerably increased the performances of such systems. By definition, a BCI is a device that enables communication without movement. For a few years, research has allowed the integration of such BCIs in games, to augment interactivity of healthy users. BCI technology has also offered new communication possibilities to severely disabled people, by enabling them to move their mouse or type an email just by thought. The non-invasiveness of EEG signals represents the major advantage of this technology.

[0017] However, EEG signals are known to be very noisy implying a very low Signal-to-Noise Ratio (SNR) and, consequently, a low information transfer rate. This low bit-rate leads to the difficulty to send complex commands and the users are rather limited to very high-level commands. Also, the consequence of this low quality signal is the slowness and the lack of reliability of some BCI- based applications, as reported by B. Obermaier et Al . in "Information transfer rate in a five-classes brain-computer interface," Neural Systems and Rehabilitation Engineering, IEEE Transactions on, vol. 9, no. 3, pp. 283-288, 2001.

[0018] Recently, researchers have started to develop new possibilities and combinations between standard BCI and BNCI, i.e. Brain/Neuronal Computer Interfaces. Unlike real BCI's, BNCI's rely on indirect measures of brain activity characterized by a better SNR. Thereby, sensors reflecting activity from the eyes (EOG) , heart (ECG) or muscles (EMG) are used as inputs. Such possibilities are reported by A. B. Usakli et Al . in "A hybrid platform based on eog and eeg signals to restore communication for patients afflicted with progressive motor neuron diseases," in Conference Proceedings of the IEEE Engineering in Medicine and Biology Society, 2009. The main interest of those signals is their abilities to provide more reliable and faster interactions between the user and the machine.

[0019] For decades, neuroscientists have studied the brain activity related to movements. They have shown that precise movements like grasping are directly controlled by the brain. In recent experiments with monkeys, it was demonstrated that a mathematical link exists between reaching and grasping movement characteristics (direction, speed) and the electrical signals recorded by electrodes implanted in the motor cortex as described by J. M. Carmena et Al . in "Learning to control a brainmachine interface for reaching and grasping by primates," PLoS Biol, vol. 1, no. 2, p. e42, 10 2003.

[0020] It is now established that locomotion differs from this scheme and is actually governed by a hierarchical system. At the lowest level of this system are found the Central Pattern Generators (CPGs) . Studies with cats have revealed that the walk generation pattern is generated by those CPGs, located in the spinal cord.

[0021] A CPG is composed of motoneurons linked together that can generate periodic patterns whose frequency is controlled by the brain. To prove this concept, scientists have sent impulsive periodic signals in a specific area in the brain stem called Mesencephalic Locomotor Region (MLR) (A. J. Ijspeert, "Central pattern generators for locomotion control in animals and robots: a review," Neural Networks, vol. 21, no. 4, pp. 642-653, 2008) . They found that the frequency of this stimulation signal determined the speed of cat's walk. By increasing the stimulation frequency, they could even make the cat trotting instead of walking.

[0022] This mechanism has inspired the field of robotics, particularly in the development of small autonomous walking robots and prostheses (G. C. Nandi, A. J. Ijspeert, and A. Nandi, "Biologically inspired CPG based above knee active prosthesis," in 2008 IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS, 2008, pp. 2368- 2373) .

[0023] One of the algorithms developed in this framework is called Programmable Central Pattern Generator (PCPG) (L. Righetti et Al . , "From dynamic hebbian learning for oscillators to adaptive central pattern generators," in Proceedings of 3rd International Symposium on Adaptive Motion in Animals and Machines - AMAM 2005. Verlag ISLE, Ilmenau, 2005) . A PCPG algorithm is able to generate any periodic pattern after a learning step.

[0024] The interest of such a system lies in the controllable aspect of the learned parameters. Actually, the pattern magnitude and frequency are easily adjustable. A modification of one of these parameters will lead to a smooth transition of the PCPG output. This is a particularly interesting feature, which is especially important for prosthesis (or orthosis) applications and their actuators.

Summary of the Invention

[0025] The present invention is related to a method for determining an artificial periodic patterned signal comprising the steps of:

providing a set of commands for driving a central pattern generator;

providing a brain computer interface (BCI); - determining a set of signals originating from said BCI, said set of signals corresponding to said set of commands ;

determining from a BCI measurement a signal pertaining to said set of signals;

- sending a command corresponding to the determined signal to the central pattern generator,

said central pattern generator producing an artificial periodic patterned signal based on said command.

[0026] In the present invention, a BCI should be understood as any interface able to determine the intent of a user without noticeable movements, including for example BCI based on EEG signal (non-invasive BCI), electro cortical signals (invasive BCI) or eye movement detection signals . [0027] In the present invention eye movement should be understood in its broadest sense, including eye-related movements such as eye blinking.

[0028] According to particular preferred embodiments, the method of the present invention comprises one or a suitable combination of at least two of the following features:

- the central pattern generator is a programmable central pattern generator;

- the method further comprises the step of training the programmable central pattern generator by using a predetermined standard walking pattern, preferably obtained by direct measurement of a real walking pattern;

- the method further comprises the step of determining the frequency and amplitude of the trained PCPG corresponding to a predetermined set of walking speeds ;

- the BCI is based on eye movement measurement signal, the commands being preferably based on eye movement sequence detection;

- the set of signals is corresponding to a set of eye movement sequence, preferably measured by electrooculography;

- the method further comprises the step of calibrating the eye movement measurement for determining the angular resolution of the eye movement measurement;

- the determination of the eye movement sequence from eye movement measurement signal comprises the step of band-pass frequency filtering the eye movement measurement signal, the band-pass filtering presenting preferably a pass frequency comprised between 0,05 Hz and 20 Hz; - the determination of eye movement sequence from eye movement measurement signal comprises the step of determining the derivative of the eye movement measurement signal (optionally filtered) ;

- the determination of eye movement sequence from eye movement measurement signal comprises the step of thresholding the derivative of the eye movement measurement signal;

- the artificial periodic patterned signal is related to a limb movement or an electromyographic (EMG) signal corresponding to said limb movement;

- said set of command comprises the command of accelerating, decelerating and stopping, the resulting speed being preferably continuously adapted by interpolation of frequency and/or amplitude between walking speeds pertaining to the predetermined set of walking speeds ;

- more than one CPG are used for generating different walking speed;

- the commands are sent simultaneously to more than one programmable central pattern generators for determining movement of a limb comprising more than one degree of freedom, said programmable central pattern generators being preferably coupled and synchronised by said coupling;

- the method further comprises the step of measuring a feature of the resulting movement for adapting the CPG to external perturbation, said adaptation being performed by a feedback loop;

- the determination of the signal pertaining to said set of signals comprises the determination of an incertitude level, said incertitude level being used to refine the command sent to the programmable central pattern generator ( s ) .

[0029] A second aspect of the invention is related to a computer readable medium having computer readable program code embodied therein for determining a periodic patterned signal, the computer readable code comprising instructions which when executed by a processor execute the method of the invention.

[0030] A third aspect of the invention is related to a system for determining an artificial periodic patterned signal comprising:

- a BCI for measuring a BCI signal;

- a central pattern generator connected to said BCI, said central pattern generator producing, in use, artificial periodic patterned signal based on said BCI signal .

[0031] The CPG used in the system is either hardware implemented or in the form of a program code on a computer readable medium.

[0032] Preferably, the CPG is a programmable CPG, in order to easily adapt to experimental gait parameters.

[0033] Advantageously, the BCI is based upon eye movement measurement signal.

[0034] Preferably, the means for measuring eye movements comprise an electrooculograph, for measuring the resting potential of the retina, the eye movement signal being determined from said resting potential.

[0035] Advantageously, the system of the invention further comprises a high-pass frequency filter able to reduce eye movement signal drifting. Preferably, said high- pass frequency filter is adapted to suppress signal below 0, 05Hz .

[0036] Preferably, the system of the invention further comprises a low-pass frequency filter able to reduce high frequency eye movement signal noise. Advantageously, said low-pass filter has a cutoff frequency of 20 Hz.

[0037] The present invention is also related to lower limb prosthesis or orthosis comprising a system according to the invention and further comprising actuators, said actuators being able to produce prosthesis or orthosis movements based on the artificial periodic patterned signal produced by the programmable central pattern generator.

[0038] Preferably, said prosthesis or orthosis movement is related to bipedal locomotion.

[0039] Advantageously, the prosthesis or orthosis comprises at least one position and/or pressure sensor used in a feedback loop of the programmable central pattern generator ( PCPG) , more particularly for synchronising said PCPG with the other limb movements.

Brief Description of the Drawings

[0040] Fig. 1 represents an example of EOG-based wearable eye tracker used in the invention.

[0041] Fig. 2 represents an example of positioning of the EOG electrodes on user's face.

[0042] Fig. 3 represents the EOG pipeline alowing the determination of the eye movement magnitude on both vertical and horizontal channels. (a) Raw EOG data, (b) Filtered EOG data, (c) derivative of the EOG data, high values indicating saccades, (d) thresholded derivative of the EOG, (e) linear regression based on the thresholded derivatives and the measured angles.

[0043] Fig.4 represents the accuracy of the obtained eye angle measurement.

[0044] Fig.5 represents the general principle of the

PCPG, based on multiple oscillators. [0045] Fig. 6 represents an example of PCPG output compared to a standard pattern of walk, using 7 oscillators .

[0046] Fig. 7 represents the output of the PCPG of the example with variable speed, with variable amplitude and frequency.

[0047] Fig. 8 represents a foot elevator orthosis used in the example.

[0048] Fig. 9 represents the evolution of the foot pattern frequency (a) and amplitude (b) as a function of waling speed.

Detailed Description of the Invention

[0049] The present invention is related to a system combining the ease of control of human/computer interface (BCI) with the use of artificial central pattern generator (CPG) .

[0050] Several types of BCI can be used, as far as they provide the possibility of generating high level commands at high speed and sufficient level of confidence. For example, EEG signal (non-invasive BCI) or electro cortical signals may be used. Such method for determining high-level command by BCI is for example described by J. del R. Millan in "Asynchronous Non-Invasive Brain-Actuated Control of an Intelligent Wheelchair", EMBC 2009, which is incorporated herewith by reference.

[0051] Advantageously, the BCI uses eye movement detection as input signal.

[0052] Preferably, said eye movement detection is based on EOG measurement.

[0053] The output of the artificial central pattern generator may advantageously be used to control movements related to human locomotion. For example, the system of the invention may be included in an active lower limb prosthesis or orthosis comprising actuators, said actuators being controlled by the output of the artificial CPG.

[0054] Preferably, the artificial central pattern generator is a programmable central pattern generator (PCPG) , so that it may easily be adapted to particular individual, gait or speed.

[0055] Advantageously, the PCPG is continuously adapted to any walking speed by continuously adapting amplitude and/or frequency of the generated periodic movement. This continuous adaptation is obtained by interpolation between measured values of the walking speed.

[0056] Preferably, the method of the invention uses more than one PCPG, for adapting the periodic pattern to different gaits. For example, such different gaits correspond to speed changes, one PCPG being used for low speed walking and another PCPG being used for higher speed.

[0057] Preferably, the method of the invention comprises the step of measuring at least one resulting movement parameter, in order to adapt the periodic pattern with other limb movements. This adaptation is advantageously used in a feedback loop. The measured feature may be either directly related to the PCPG generated movement, or related to a feature with which the PCPG should be synchronised. In the case of an ankle orthesis for example, the PCPG may be synchronised with hip or knee angle measurement.

Description of a Preferred Embodiment of the Invention EOG EYE MOVEMENT DETECTION

[0058] In the case of severely disabled people, eye movements are often one of the last means of communication. This is why researchers have tried to interpret eye movements in order to interact with a computer, namely a Human-Computer Interface (HCI) . Although a lot of different methods exist to track the eye movements such as special contact lenses, infrared light reflections measured with video cameras, EOG signals with electrodes around the eyes represents one of the most portable systems.

[0059] Therefore, many researchers have studied the way to achieve rehabilitation thanks to this signal. Some research was performed for handling graphical interfaces where eyes are used as a mouse controller, for controlling a robot, helping disabled people to communicate or to move a wheelchair.

[0060] An example of such portable measurement system 4 is represented in Fig. 1, wherein six dry electrodes 1-2 are fixed by means of flexible supports onto goggles 5. In such system, four electrodes 1 are disposed above and below each eyes, in order to measure the vertical movements of both eyes, and two electrodes 2 are disposed on both sides of the face, in order to measure lateral displacements. Optionally, an additional electrode could be used between the eyes.

[0061] A light sensor 3 is advantageously added between the eyes for compensating EOG signal artifacts induced by changes in ambient light intensity.

[0062] The signals originating from the different sensors/electrodes are directed by means of conductors 6 (only three are represented on fig.l for the sake of clarity) to an electronic device 7 for treating the different signals. Advantageously, said electronic device comprises means for amplifying the signals (pre-amplifier) and a digital signal processing unit (DSP) for real-time EOG signal processing.

[0063] Preferably, two accelerometers 8 measuring rotations of the head along two axes are fixed on the goggles 4 for compensating EOG signal artifacts caused by physical activity. The relevant rotation axes ACCy and ACCz relative to the user's head are represented on Fig. 2.

[0064] Advantageously, the DSP comprises means for streaming the processed EOG signals to a remote device over communication means such as Bluetooth to drive other systems .

[0065] The electrodes measure the resting potential that is generated by the positive cornea (front of the eye) and negative retina (back of the eye) . When the eye rotates, the dipole rotates as well. By positioning pairs of electrodes around the eyes as shown in Figure 2, it is possible to decompose the eye movements on the horizontal and vertical axes. This technique has the advantages of a very high signal to noise ratio (SNR) while being easily portable and cheap. Such EOG device is described by A. Bulling, "Eye movement analysis for context inference and cognitive-awareness: Wearable sensing and activity recognition using electrooculography, " Ph.D. dissertation, ETH Zurich, 2010, which is incorporated hereby by reference.

[0066] Eye movements can be classified as following: fixations, saccades and eye blinks.

[0067] Fixations are the stationary states of the eyes during which gaze is focusing on a particular point on the screen. Saccades are very quick eye movements between two fixations points. The duration of a saccade depends on the angular distance the eyes travel during this movement. For a distance of 20 degrees, the duration is between 10 ms and 100 ms . Eye blinks cause a huge variation in the potential in the vertical electrodes around the eyes. Those movement types, which last between 100 ms and 400 ms, can be used to control an HCI .

[0068] Preferably, saccade detection is used to construct the eye-tracker. Figure 3 illustrates the pipeline executed for detecting saccades present in both the vertical and horizontal EOG signals:

1) High-pass filter (0.05 Hz) for drift correction which is very strong in the EOG signals.

2) Low-pass filter (20 Hz) to reduce high frequency noise without affecting the eye movements.

3) Derivative in order to detect the rapid variations .

4) Thresholding to detect saccades and remove noise and integration in the saccade range, which defines the EOG feature.

5) Linear regression between the angle and the integration result (EOG feature) .

6) Conversion to x,y position.

[0069] The offline analysis protocol was twofold. In order to get enough data for training the linear regression, 25 trials were used. In each trial, the 4 subjects were asked to look at a target jumping alternatively from the centre of the screen to one of the five following possibilities: extreme top, bottom, left, right and centre targets. For horizontal and vertical eye movements, there were separate pipelines and the regression was also trained separately.

[0070] For evaluation 100 trials were assessed. A method was developed in order to determine the angular resolution achievable for each subject. The angular resolution is defined as the limit angle under which two gaze directions are not distinguishable anymore using the EOG signals. In practice, this value was taken as the accuracy of the system at N centimetres maximum deviation from the target. The screen was divided in a 5 by 5 grid, resulting in 25 potential target positions, which were selected randomly. The jump between the centre and the target of each trial was considered correct when the Euclidean distance between the EOG-based estimation point on the screen and the actual point was lower than N centimetres. The angular resolution was then obtained by looking at the wanted precision, typically 95% or 99% and by converting centimetres on the screen into angles.

[0071] The data were recorded using the BioSemi

ActiveTwo hardware, with flat active electrodes positioned according to Figure 1. The distance between the user and the screen was 70 cm. The result for one subject is exposed in Figure 4.

[0072] Although similar results can be derived from vertical EOG signals, a template-based approach for eye blink detection can be developed to enhance the performance as described by A. Bulling. The obtained template of eye movements is then modelled by a two Gaussian distribution. Based on Euclidean distance, this pattern is matched to the EOG signals to detect eyeblinks and correct them or use them as a complementary input.

HUMAN WALK MODELED BY A PCPG

[0073] Interestingly, different gaits of animals can be simulated and reproduced with diverse walking robots by means of CPG. Contrary to usual CPG oscillators, the learning of Programmable Central Pattern Generators (PCPGs) is very easy and avoids challenging and heavy parameterisation .

[0074] Furthermore, this oscillating system is able to change the frequency and magnitude of any given periodic walking pattern it has learned in a smooth way and is robust to noise and to perturbations.

[0075] Preferably, supervised learning techniques are used in order to determine the CPG parameters. In such techniques, the desired rhythmic pattern that the CPG should produce is known (target pattern) . The desired pattern can then be used to define an explicit error function to be minimized.

[0076] Examples of learning techniques include (but are, not limited to) :

gradient-descent learning algorithms for recurrent neural networks as described for example by Prentice et Al . in "Simple artificial neural network models can generate basic muscle activity patterns for human locomotion at different speeds", Experimental Brain

Research, 123, 474-480;

- learning for vector fields as described by Okada et Al . in Polynomial design of the nonlinear dynamics for the brainlike information processing of whole body motion. In IEEE International Conference on Robotics and Automation (ICRA2002) 1410-1415;

- statistical learning algorithms (e.g. locally weighted regression) for dynamical systems as described by Nakanishi et Al . in Learning from demonstration and adaptation of biped locomotion. Robotics and

Autonomous Systems, 47, 79-91

- programmable central pattern generators that use pools of frequency adaptive oscillators to learn a specific rhythmic pattern as described by Righetti et Al . in "Programmable central pattern generators: an application to biped locomotion control, Proceedings of the 2006 IEEE International Conference on Robotics and Automation (ICRA2006) (pp. 1585-1590) . An interesting aspect in this learning procedure is that learning is embedded into the dynamical system.

[0077] A PCPG is a kind of Fourier series decomposition and is composed of several adaptive oscillators. Preferably, the CPG algorithm is governed by the following equation system:

with

and

[0078] As depicted in Figure 5, oscillators are coupled between each other compared to an origin phase based on the Ri coupling parameters. They are composed of adaptive magnitude coefficient and frequency parameters

[0079] The signal resulting from the sum of

oscillator outputs is compared to the walking

pattern target and the error value F(t) is computed. Throughout the learning step, all the parameters of the PCPG are modified in order to minimize F(t) . When this learning step is finished, F(t) is close to zero and the system is generating the right pattern at the

output.

[0080] In this section, we demonstrate that human walk can be learned by a PCPG and subsequently generated at different walking speeds. In order to train the PCPG, three standard walking patterns were used. These temporal patterns consist of the angle of elevation of the foot, the thigh and the shank of a healthy subject walking on a treadmill at 3 km/h, a typically medium speed for humans. The elevation angles were computed using the positions of 23 passive markers disposed on the subject, determined thanks to six Infrared Bonita Vicon cameras. [0081] The standard walking patterns were obtained by averaging about 50 walking cycles, determined with a peak detection algorithm able to locate all the maximum angle values of the kinematics recordings. The kinematics data were recorded for each leg during 40 seconds at 100 Hz. Each standard pattern is thus a kind of average pattern along the 40second recordings.

[0082] After determining these standard patterns, the PCPG was trained using the procedure described by L. Righetti et Al . in "From dynamic hebbian learning for oscillators to adaptive central pattern generators," in Proceedings of 3rd International Symposium on Adaptive Motion in Animals and Machines - AMAM 2005 Verlag ISLE, Ilmenau, 2005, conference, which is hereby incorporated by reference.

[0083] Properties of PCPGs make them suitable for trajectory generation in robotics and also for prosthesis applications. In fact, the pattern learned by a PCPG can be easily controlled in magnitude and in frequency thanks to a simple linear change of the vectors representing the R^N PCPG states (N is the number of oscillators) . This linearity leads to a smooth change of the global system behavior. Figure 7 depicts the various modifications relevant for prosthesis control purposes. For instance, if the vector is divided by two, the underlying frequency of the standard temporal pattern is divided by two. The same effect occurs for the vector.

[0084] Advantageously, in order to adapt the CPG to real walking pattern, sensors are used to synchronise the CPG with the other limb movements. This adaptation have the advantage of coupling the CPG with the natural CPGs of a user in a walking process. [0085] At constant speed, gait cycles are usually not perfectly identical. This fact and numerous perturbations can induce phase mismatch between the perfectly periodic CPG output and the real walking pattern. If this mismatch is too important, the

subject has to compensate for it leading to a non-natural walk. The aim of this phase resetting is to pave the way to allow the orthosis to adapt to the patient as quickly and smoothly as possible aiming at increasing the subject comfort.

[0086] Advantageously the phase resetting consists in resynchronizing the PCPG state according to special events. Therefore, the PCPG will be phase reset on the HS to allow the system to recover the correct phase in a smooth way at the time of the T₀. Two approaches are available: a hard and a soft phase resetting.

[0087] The hard phase resetting relies on a direct modification of the integrated values: in each oscillator i, Xi and i are put to standard values corresponding to the HS event. The main advantage of this approach is the quick phase-locking whereas the disadvantages are a more sensitive reaction to noise in the measurement itself and perturbations due to small variations in gait cycles at constant speeds or and instability of the user because of rapid modifications. In the case of a foot lifter orthosis, during the stance phase, the actuator is not commanding the system and thus, the latter disadvantage is mitigated.

[0088] However, it is a real problem in the case of a complete prosthesis.

[0089] In the soft phase resetting, the original PCPG algorithm is slightly modified. To control the phase of the first CPG oscillator, a coupling with a reference oscillator at instantaneous phase Ro, was established. This allows to modify the phase difference between the

reference oscillator and the first oscillator of the PCPG. Formally, the reference oscillator is as follows:

whereas the coupling with the PCPG (subscript by p) is shown in:

where . The coupling with the other oscillators of

the PCPG is identical to the previous description. Because the phase of higher order oscillators have more difficulties to follow a phase change in experiments, coupling constant was defined as is then used

instead of τ in the oscillators of order i of the PCPG (pcpg.x (i,p) )

WHOLE SCHEMATIC CONTROL

[0090] The present invention have been used in a biologically inspired process to control a lower limb prosthesis 10 by BCI based signal (including EOG) as depicted in Figure 8. The experimental process is composed of a high-level command system based on EOG signals and a pattern generation to control this prosthesis (or orthosis) .

[0091] The orthosis 10 of this example is made of several components: a custom-fit plastic shell for the shank 11, and another plastic sheel for the foot 13 a flexure joints 12, a linear actuator 19 fixed to the shank shell 11 by means of fastening means 20, a ball-link transmission 16, a load cell 17 to measure the actuator force, and two force sensors 14,15 installed in the orthosis sole, under the heel and the toes. The plastic shells 11,13 were designed using a 3D scan of the right foot and leg of a healthy subject, adding mounting surfaces for the actuator, the flexure joints, and the mechanical transmission 18. The actuator includes a position control unit that can be driven by an external analog signal in the range of 0 to 10 V.

[0092] Utilizing the EOG output presents a double interest. First, eye gaze detection can provide the precise direction of eye movements in real time. These movements can thus be labelled as left, right, up or down.

[0093] Specific eye movement sequences executed by the user can then be associated to high-level commands sent to the prosthesis actuator. It appears that a simple and quite effective mean to determine eye movement sequences is to use the edit (or Levenshtein) distance.

[0094] The Levenshtein distance between two given strings is defined as the number of deletions, insertions and substitutions required to transform one of them into the other one. In this case, the string is built by the concatenation of each labelled state of the eyes (e.g. the string associated to a left-right movement would be LR) .

[0095] A second interest of EOG signals resides in the high speed of eye movements. The user can thus very quickly activate or deactivate a high-level command generation environment.

[0096] This aims at decreasing the subject attention load to actually command the prosthesis. For example, when the patient wants to change the speed, he enters in this environment by means of a certain sequence of blinks and executes the correct eye sequence to really change the speed .

[0097] In case of emergency, the patient could stop the prosthesis using a specific high-level command (a sequence of winks for example) recognizable without entering in the environment mode. From a practical point of view, it has to be mentioned that an adaptative filter has been developed to remove artifacts due to walk limiting therefore the error rate in eye movement sequences detection.

[0098] What is proposed in this invention is to generate walking patterns with the PCPG in a way differing from the bipedal robots described in the literature. Indeed, one of the main goals in prosthetics is to provide the user with the most comfortable walk possible.

[0099] Therefore, at each step, the pattern should be adapted in terms of frequency and magnitude, i.e. respectively the stepping frequency and stride-related length between two heel strikes whatever the walking speed. Kinematics data were thus recorded with the same subject and apparatus for 10 different speeds, from 1.5 to 6 km/h, by step of 0.5 km/h.

[0100] The pattern learned and generated by the PCPG for the speed of 3 km/h was manually calibrated (by tuning the magnitude and frequency parameters) in order to fit the standard walking patterns of all the other speeds. By this procedure, we found a mathematical link between the PCPG amplitude and frequency parameters (the vectors) as a function of the walking speed. Figure 9 shows results obtained for one subject. A four-degree polynomial function seems to be sufficient to model the evolution of PCPG parameters versus the ten recorded walking speeds.

[0101] This polynomial fit permits to interpolate any walking speed between the extreme learned values, giving very high comfort level to the user.

[0102] One can notice that the subject increases his walking speed at first by extending his stride length, and then by increasing his stepping frequency. It has to be emphasized that this interpolation can be computed specifically for any subject, increasing therefore the precision and adequacy of the prosthesis control at each step .

[0103] In some cases, it was observed that a gait change occurs at particular walking speed. In that case, the use of more than one CPG may improve the quality of the determined periodic pattern corresponding to more natural pattern .

CONCLUSION

[0104] As a summary, the present invention discloses method for determining a periodic or quasi-periodic movement based on EOG signal (or high-level BCI) . The disclosed method have been shown to be adapted to drive a lower limb prosthesis. This method is composed of two main steps. At first, an EOG-based eye tracking system generates high-level commands (faster, slower, stop, ...) on the basis of specific eye movement sequences executed by the user, and then, the determined command is used to determine PCPG parameters such as speed and/or amplitude by the mapping between PCPG parameters and real walk patterns.

[0105] After learning average walking patterns (angles of elevation of the different parts of the leg as a function of time) , a PCPG provides an adaptive kinematics output to drive the artificial limb, according to the walking speed desired by the user. Unlike current sophisticated active prostheses, the user's intent is fully taken into account in this case.

[0106] A method to process raw EOG signals in order to detect eye movement was also described. In addition, a method to determine, for a given subject, the minimum angular resolution achievable with his or her EOG signals is proposed.

[0107] It was also demonstrated that a PCPG is able to learn almost perfectly average human walk patterns. Moreover, it is shown that a four-degree polynomial function can model the evolution of the PCPG parameters as a function of the walking speed. This interpolation enables to drive the prosthesis in a smooth way during accelerations or decelerations, increasing thus the comfort of the patient.

[0108] Preferably, the recognition of eye movements sequences comprises the step of evaluating confidence level in this recognition procedure and said confidence level is integrated in the control system itself. For instance, if the decision to increase the speed is sure at 75 %, 75 % of the speed increase is actually performed.

[0109] In the example, emphasis was given to a prosthesis (or orthosis) with only one joint controlled by one PCPG (ankle) . Nevertheless, the proposed principles are easily generalisable to more complex prosthesis, for example for above-knee amputees, with prosthesis having more than one joint to be controlled, and having coupled PCPG' s as described for example by Florian Hackenberger in his thesis "Balancing Central Pattern Generator based Humanoid Robot Gait using Reinforcement Learning". In this case, one would use one PCPG for each joint and a coupling between the PCPG in order to maintain synchronisation.

[0110] The man skilled in the art would also easily recognise that the generated PCPG signal may be used by a shaping neural network leading to EMG signals. These signals could be the input of a Functional Electrical Stimulation device (FES) .This type of device may be useful in case of disabled patient having still their limbs but having nerves dysfunction such as disrupted spinal cord. (See for example the article of S. D. Prentice et Al . in "Artificial neural network model for the generation of muscle activation patterns for human locomotion", Journal of Electromyography and Kinesiology 11 (2001) 19-30 and S. D. Prentice, A. E. Patla, and D. A. Stacey, "Simple artificial neural network models can generate basic muscle activity patterns for human locomotion at different speeds," Experimental Brain Research, vol.123, pp. 474-480, 1998) .

Claims

1. Method for determining an artificial periodic patterned signal comprising the steps of:

providing a set of commands for driving a central pattern generator;

providing a brain computer interface;

determining a set of signals originating from said brain computer interface, said set of signals corresponding to said set of commands;

determining from a brain computer interface measurement a signal pertaining to said set of signals;

sending a command corresponding to the determined signal to the central pattern generator,

2. Method according to claim 1 wherein the artificial periodic patterned signal is related to a limb movement or an electromyographic signal corresponding to said limb movement.

3. Method according to any of claims 1 or 2 wherein the central pattern generator is a programmable central pattern generator.

4. Method according to claim 3 wherein the predetermined artificial periodic pattern signal is corresponding to walking movement, and the method further comprising the step of training the programmable central pattern generator by using predetermined standard walking pattern .

5. Method according to claim 4 wherein the standard walking pattern is obtained by direct measurement of a real walking pattern.

6. Method according to any of claims 4 or 5 wherein said set of command comprises the command of accelerating, decelerating and stopping.

7. Method according to claim 6 wherein the resulting speed being continuously adapted by interpolation of frequency and/or amplitude between learned values.

8. Method according to any of the previous claims wherein more than one central pattern generator are used for generating different artificial periodic pattern signal.

9. Method according to any of the previous claims further comprising the step of measuring a feature of a movement resulting from said artificial periodic pattern for adapting the central pattern generator to external perturbation, said adaptation being performed by a feedback loop.

10. Method according to any of the previous claims wherein the commands are sent simultaneously to more than one programmable central pattern generators for determining movement of a limb comprising more than one degree of freedom, said programmable central pattern generators being coupled and synchronised by said coupling.

11. Method according to any of the previous claims wherein the BCI is based on eye movement measurement signal.

12. Method according to claim 11 wherein the set of signals is corresponding to a set of eye movement sequences .

13. Method according to claim 12 wherein the determination of the eye movement sequence from eye movement measurement signal comprises the step of band-pass frequency filtering the eye movement measurement signal, the band-pass filtering presenting preferably a pass frequency comprised between 0,05 Hz and 20 Hz.

14. Method according to any of claims 12 or

13 wherein the determination of eye movement sequence from eye movement measurement signal comprises the step of determining the derivative of the eye movement measurement signal.

15. Method according to any of claims 12 to

14 wherein the determination of eye movement sequences from eye movement measurement signal comprises the step of thresholding the derivative of the eye movement measurement signal.

16. Method according to any of claims 11 to

15 wherein the eye movements are measured by electrooculography .

17. Method according to any of claims 11 to 16 further comprising the step of calibrating the eye movement measurement for determining the angular resolution of the eye movement measurement.

18. Method according to any of the previous claims wherein the determination of the signal pertaining to said set of signals comprises the determination of an incertitude level, said incertitude level being used to refine the command sent to the programmable central pattern generator .

19. Computer readable medium having computer readable program code embodied therein for determining a periodic patterned signal, the computer readable code comprising instructions which when executed by a processor execute the method according to any of the previous claims.

20. System for determining an artificial periodic patterned signal comprising:

- a brain computer interface for measuring a BCI signal;

- at least one central pattern generator connected to said brain computer interface, said central pattern generator producing, in use, artificial periodic patterned signal based on said brain computer interface signal.

21. System according to claim 20 wherein the central pattern generator is a programmable central pattern generator, in order to easily adapt to experimental gait parameters .

22. System according to claim 20 or 21 wherein the brain computer interface comprises an electrooculograph .

23. Lower limb prosthesis or orthosis comprising a system according any of claims 20 to 22 and further comprising actuators, said actuators being able to produce prosthesis or orthosis movement based on the artificial periodic patterned signal produced by the central pattern generator.

24. Lower limb prosthesis or orthosis according to claim 23 wherein the movement is related to bipedal locomotion.

25. Lower limb prosthesis or orthosis according to claim 23 or 24 comprising at least one position and/or pressure sensor used in a feedback loop of the central pattern generator.