User interventions

Abstract

An apparatus comprising a body portion 101 mountable to a body part of a user, a controller 103 and a set of modulation devices 105 provided on the body portion, each device 107 of the set configured

Description

FIELD The present invention relates generally to apparatus and machine-readable storage media for regulating deployment of interventions for a user. BACKGROUND In general, deviations in the homeostasis of the human body may be caused by extrinsic and / or intrinsic factors that can disrupt the equilibrium between different physiological processes. Extrinsic factors can include, for example, drug / substance abuse, circadian rhythm changes (e.g., jetlag), exposure to radiation, stress / anxiety, traumatic injury and so on, whilst intrinsic factors can include, for example, genetic mutations resulting in disease development, congenital diseases, autoimmune disorders (e.g. multiple sclerosis), neurovascular diseases (e.g. stroke), musculoskeletal pathologies (dyskinesias, joint disorders) and so on. A tremor is an example of a manifestation of underlying distortions to human body homeostasis and comprises an unintentional and uncontrollable rhythmic movement of one part or a limb of the body. Tremors can be intermittent or constant, and they can develop on their own or indicate an underlying health issue such as a neurological disorder, neurodegenerative disease or deficiency. The underlying cause can be determined by, e.g., CT or MRI imaging to investigate structural defects and / or degeneration of the brain, and / or blood / urine tests to investigate any deficiencies or metabolic causes. Such diagnostic solutions are typically separate from any treatment device or methodology. Aside from tremors, there are numerous underlying distortions to human body homeostasis that result in unwanted symptoms / manifestations. Attempts at restoring normalcy to a patient’s everyday life can fall short of offering a compelling solution set that captures the needs of the patient, caregiver and attending physician. SUMMARY According to a first aspect of the present disclosure, there is provided an apparatus, comprising a body portion mountable to a body part of a user, a controller, and a set of modulation devices provided on the body portion, each modulation device of the set of modulation devices configured to receive a respective control signal from the controller to regulate a function thereof, wherein the controller is configured to receive sensor data from a set of sensors, the data representing at least one of a set of environmental data and a set of user data, using the sensor data, determine a set of operational parameters for the set of modulation devices, the set of operational parameters relating to a hybrid stimulatory output of the set of modulation devices, and using the set of operational parameters, generate the set of control signals. In an implementation of the first aspect, at least one of the set of modulation devices can comprise an electrical neuromuscular stimulator. At least one of the set of modulation devices can comprise an electrical heating element. At least one of the set of modulation devices can comprise a Peltier device. At least one of the set of modulation devices can comprise a gyroscopic device. In an example, at least one of the set of modulation devices can comprise a device configured to deliver at least one of a visual, and an acoustic signal. At least one of the set of modulation devices can comprise a device configured to deliver a mechanical stimulation. The set of modulation devices can be configured to deliver sensory stimulation for the user. At least one of the set of sensors can comprise an inertial measurement unit or a force sensor. At least one of the set of sensors can be user mountable to the body portion. At least one user mountable sensor can be implantable or embeddable in the body portion. The at least one user mountable sensor can be configured to generate the set of user data representing one or more characteristics or vital signs of the user. At least one of the set of sensors can be configured to generate the set of environmental data representing a state of at least one environmental parameter. The apparatus can further comprise a transceiver configured to receive at least one of the set of environmental data and the set of user data. The set of modulation devices can be configured to apply the hybrid stimulatory output using the set of control signals. According to a second aspect of the present disclosure, there is provided a machine-readable storage medium encoded with instructions for controlling a set of modulation devices of an apparatus, the instructions executable by a processor of the apparatus, whereby to cause the apparatus to receive sensor data from a set of sensors, the data representing at least one of a set of environmental data and a set of user data, using the sensor data, determine a set of operational parameters for the set of modulation devices, the set of operational parameters relating to a hybrid stimulatory output of the set of modulation devices, and using the set of operational parameters, generate the set of control signals to control respective functions of the modulation devices to generate the hybrid stimulatory output. In an implementation of the second aspect, the machine-readable storage medium can comprise instructions to cause the apparatus to, using the sensor data, determine a set of parameters representing a current state space for the user, determine a set of actions on the basis of the current state space for the user, using a policy function, determining the set of control signals, wherein the set of control signals are configured to cause the hybrid stimulatory output of the set of modulation devices to be executed, thereby affecting the set of actions and thereby modifying the state space of the user. The machine-readable storage medium can comprise instructions to cause the apparatus to regulate an intensity of the respective functions of the modulation devices. The machine-readable storage medium can comprise instructions to cause the apparatus to regulate an operational state of one or more of the modulation devices. The machine-readable storage medium can comprise instructions to cause the apparatus to generate a set of operational parameters on the basis of a position of at least one of the modulation devices. The machine-readable storage medium can comprise instructions to cause the apparatus to modify a predetermined hybrid stimulatory output on the basis of the sensor data to generate a modified hybrid stimulatory output. The machine-readable storage medium can comprise instructions to cause the apparatus to generate a set of control signals for the modified hybrid stimulatory output. BRIEF DESCRIPTION OF THE FIGURES Embodiments of the invention will now be described by way of example only with reference to the figures, in which: Figure 1 is a schematic representation of an apparatus according to an example; and Figure 2 is a schematic representation of an apparatus according to an example. DETAILED DESCRIPTION Example embodiments are described below in sufficient detail to enable those of ordinary skill in the art to embody and implement the systems and processes herein described. It is important to understand that embodiments can be provided in many alternate forms and should not be construed as limited to the examples set forth herein. Accordingly, while embodiments can be modified in various ways and take on various alternative forms, specific embodiments thereof are shown in the drawings and described in detail below as examples. There is no intent to limit the particular forms disclosed. On the contrary, all modifications, equivalents, and alternatives falling within the scope of the appended claims should be included. Elements of the example embodiments are consistently denoted by the same reference numerals throughout the drawings and detailed description where appropriate. The terminology used herein to describe embodiments is not intended to limit the scope. The articles “a,” “an,” and “the” are singular in that they have a single referent, however the use of the singular form in the present document should not preclude the presence of more than one referent. In other words, elements referred to in the singular can number one or more, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises,” “comprising,” “includes,” and / or “including,” when used herein, specify the presence of stated features, items, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, items, steps, operations, elements, components, and / or groups thereof. Unless otherwise defined, all terms (including technical and scientific terms) used herein are to be interpreted as is customary in the art. It will be further understood that terms in common usage should also be interpreted as is customary in the relevant art and not in an idealized or overly formal sense unless expressly so defined herein. According to an example, a multi-modal (hybrid) modulation apparatus is provided. The apparatus can be in the form of a movement / physiological / tremor modulation apparatus. The apparatus can be removably mounted or positioned on a part or parts of a body of a user. The apparatus can determine physiological and / or external or extrinsic (e.g., environmental) parameters and generate a set of control signals that can be used to control a set of modulation devices in order to, e.g., elicit a response that restores human body homeostasis. In an example, the apparatus comprises a body portion and a controller. The set of modulation devices can be provided on the body portion. For example, a modulation device can be fixedly or removably mounted to the body portion. The body portion can comprise a profiled or flexible support for the modulation devices that is attachable to the part or parts of the user, such as the user’s wrist, finger(s), thumb, limbs, and so on by means of, e.g., straps, such as straps using a hook and loop-type adjustable securing arrangement. The flexible support can comprise a fabric support, for example, that can comprise a soft, comfortable material that can be worn comfortably for extended periods of time. In some examples, the fabric can be of the type described in, e.g., WO 2014 / 127291 in which van der Waals forces are developed between a soft silicone fabric surface and a wearer’s skin, thereby retaining the fabric in place. In some examples, a flexible support can comprise an elastomeric or polymeric material, or any other material with a selected physical morphology. For example, a metallic support profile using, e.g., Kirigami techniques, or a chainmail-type structure could be used. In some examples, the body portion can attach to a body part of a user by way of, e.g., adhesive or suction means. In some examples, a body portion, or part thereof, can comprise a rigid member, such as a rigid member that is so profiled or formed as to match the profile or form of a body part. The rigid member can be formed from, e.g., a plastic material, or any other suitable formable rigid material. In some examples, the body portion can comprise a set of modulation devices so spaced from one another as to enable separate parts of a user’s body to be stimulated in a predetermined manner, as will be described in more detail below. So, for example, the body portion may extend over a user’s arm with modulation devices so positioned as to enable a predetermined stimulus to be applied or presented at the distal and proximal regions of the arm. According to an example, each modulation device of the set of modulation devices is configured to receive a respective control signal from the controller to regulate a function thereof. A modulation device can comprise an active device configured to respond to a control signal. Regulation of the function of a modulation device can comprise whether the device is on or off, a degree to which the function is presented to a user, e.g., the intensity, timing of the function (such as on / off timing, timings for variations in intensity, and so on). In some examples, a modulation device can comprise a passive device. Such a passive device can present an inherent response proportionality. For example, resistive elastic members, mass dampers, rotational masses or fluidic dampeners etc. can be provided which can be always on and self-modulating. A modulation device can provide any one or more of an electrical, thermal, vibrational, mechanical, acoustic, or visual modulation mechanism. Other modulation mechanisms can also be implemented. For example, a modulation mechanism comprising, e.g., a prompt or reminder for a user to engage in a mindfulness intervention, such as meditation for example, can be provided. Such interventions have been shown to have a damping effect on, e.g., tremors. For example, a modulation device can comprise an electrical nerve stimulator, a heating element, such as a resistive heating element for example, a Peltier device, a gyroscopic device, or a device configured to deliver at least one of a visual, or an acoustic signal for example. Each modulation device is configured to deliver a sensory stimulation for the user according to its functional / modulative modality. In an example, a modulation device can comprise multiple collocated devices. For example, a device can comprise a combined visual and acoustic device that is configured to present one or both of a visual and acoustic signal. Another exemplary multifunctional device can comprise a device that is configured to provide both heating and electrical nerve stimulation. In some examples, it can be the case that a multifunctional device is in the form of a package comprising multiple (e.g., two or more) distinct devices. In some examples, a multifunctional device can be a package comprising a single device capable of providing multiple modalities. According to an example, the set of modulation devices can comprise a suite of devices, respective ones of which are selected to provide a modality selected from those noted above for example. Thus, for example, a set of modulation devices can comprise a device to administer an electrical nerve impulse, a device to administer heating and / or cooling, a gyroscopic device to stabilise a motion, a device to deliver a visual stimulus, and a device to provide an acoustic stimulus, or a combination thereof in the case of a multifunction device. According to an example, any combination of such modulative modalities can be provided for a user, with each modulation device being individually addressable / controllable in order to enable an intensity of the function offered by the device to be regulated and / or the state of the device to be controlled. So, for example, an apparatus with a set of modulation devices comprising a heating device and an acoustic device can be controlled in order to apply one or both (or none) of these modalities for a user, at various intensity levels, at various times, and for various periods of time. The manner in which the devices are controlled can be predicated on the basis of a predetermined profile for a user, real time control or a combination of these. For example, a predetermined profile can be used to provide relief for a specific ailment of a user. It can be generic, or it can be tailored for an individual. In an example, a predetermined profile can be modified by a caregiver and / or user. In an example, a predetermined profile can be modified by way of machine learning, in which data from, e.g., sensors and / or a caregiver and / or the user can be used to modify the predetermined profile in order to provide a more tailored solution. In some examples, on the basis of a set of input data relating to a user, as will be described in more detail below, a combination of tailored modulative modalities can be provided for the user. That is, the devices can be configured to receive a set of inputs that provide a tailored response for a user based on a set of data gathered from at least one of the user and the environment. An exemplary profile for an apparatus comprising a pair of modulation devices - in this case a heating device and an acoustic device - can cause the heating device to, e.g., activate at a first intensity (i.e., temperature in this case) for a first predetermined period of time, whilst the acoustic device can activate at a first volume level with a first audible signal (e.g., a specific musical or other tonal signal for a user) for a second predetermined period of time, which may be the same or different to the first predetermined period of time. Activation of the devices may coincide, overlap or be distinct from one another. The predetermined profile can cause, e.g., the heating device to switch to a second intensity after the first predetermined period of time for a third predetermined period of time. The predetermined profile can cause, e.g., the acoustic device to change into a non-operational state of operation (i.e., off) after the second predetermined period of time. Of course, there are multiple permutations that can be envisaged for devices in terms of their functional state (e.g., on, off, intensity level, length of time of operation, time at which a function is switched on / off and so on). In some examples, an apparatus can comprise a mechanical retardation device, such as a gyroscope, in combination with a suite of devices configured to provide one or more of a mechanical stimulation (e.g., vibration), electrical stimulation (e.g., electrical neuromuscular stimulation), thermal stimulation, visual stimulation, acoustic stimulation and so on. In some examples, an apparatus can comprise a suite of devices configured to provide one or more of a mechanical stimulation (e.g., vibration), electrical stimulation (e.g., electrical neuromuscular stimulation), thermal stimulation, visual stimulation, acoustic stimulation and so on. A mechanical retardation device can be provided in the form of, e.g., a fall arrest structure, comprising e.g., a set of mechanically actuated braces configured to adjust the position of one or more of a user’s appendages in the event a fall is detected, or in the event that the precursors to a fall are detected. A functional state of a device can be based on one or more of a set of environmental and a set of user data (e.g., physiological / clinical data). According to an example, a set of sensors can be provided. The set of sensors can be used to generate sensor data representing at least one of the set of the environmental data and the set of user data. The sensor data can be used to determine a set of operational parameters for the set of modulation devices. In an example, the set of operational parameters can relate to a hybrid stimulatory output of the set of modulation devices. That is, the modulative modality of the devices can be regulated according to sensor data generated by the set of sensors. In this connection, the set of operational parameters can be used to generate the set of control signals. For example, a sensor can be provided that generates data representing movement of a user of an apparatus. The sensor may be user based (e.g., positioned on the user), or may be, e.g., a static non-user-based sensor that is configured to generate data representing the movement of a user in a particular region of an environment (for example, the static sensor may be a video camera). If the user is about to fall over for example, the sensor can generate data, such as a wavelet shaped motion data pattern that indicates a fall is about to happen. A modulation device of the apparatus can receive a control signal that is generated on the basis of this sensor data (in this case indicating that a fall is likely, is about to happen, or is happening) in order to enable the apparatus to execute a response to prevent or arrest the fall. For example, the control signal can cause a modulation device to provide biomechanical stimulation in a body part or parts that are geared to prevent the fall. In addition (or alternatively) the apparatus can trigger a warning / alarm to family and / or health care providers of the accident. Another example of a homeostasis disrupting factor is a panic attack. In this case, a sensor can be configured to generate physiological data (user data). The data can be used to detect, e.g.. increases in heart rate and / or blood cortisol levels. Such a detection can trigger generation of a control signal to effect a response from a modulation device in the form of, e.g., auditory and / or visual stimuli to calm the user and mitigate the stress response. In an example, data generated by one or more sensors and / or control signal data can be provided to and / or utilised by various stakeholders. Examples of such stakeholders include, e.g., the user, the user’s family, the user’s carer(s), healthcare professionals (e.g., specialist physicians, general physicians, nurses, therapists, pharmacists etc.), an entity that supplied the apparatus and / or sensors (e.g., for further development and R&D), insurers, those engaged in, e.g., further therapy development (e.g., medical devices, pharmacotherapy, cell therapy, physical therapy, mental therapy etc.), and so on. Figure 1 is a schematic representation of an apparatus according to an example. In the example of figure 1, a body portion 101 is depicted. The body portion 101 is mountable to a body part of a user (not shown). For example, the body portion can be in the form of a glove-shaped fabric support as described above that can be mountable (i.e., worn in this case) to a user’s hand and which may extend up the user’s wrist and that may extend up a portion of the user’s arm for example. Other ways of mounting the apparatus, or enabling a user to wear the apparatus can be provided. For example, the apparatus can be provided as part of or in the form of a smart watch, or provided on a semi-rigid structure such as a splint. In other examples, the apparatus can be provided as part of an item of clothing or footwear / headwear, or as a wearable accessory (e.g., jewellery, backpacks, glasses including VR and / or AR equipment) and so on. In some examples, a user of the apparatus can comprise a non-human user, such as a pet or other animal (e.g., a captive animal receiving care or treatment) for example. Accordingly, the apparatus can be provided as part of a jacket, collar or other structure that can be worn by or mounted on the pet. The body portion can be mountable to other parts of a user’s body. In an example, the body portion can comprise a structure that enables it to be mounted to multiple body parts simultaneously, such as multiple portions connected by fabric webbing configured to enable routing of control and / or power cabling etc. A controller 103 is provided for the apparatus, and a set of modulation devices 105 are provided on the body portion. In the example of figure 1 the controller 103 is provided on the body portion, although it may be provided separately and connected to the modulation devices by way of a wired or wireless connection (e.g., using a short-range radio frequency communication mechanism such as Bluetooth, Bluetooth LE, ZigBee and so on). In an example, each modulation device 107 of the set of modulation devices 105 is configured to receive a respective control signal 109 from the controller 103 in order to regulate a function thereof. For example, a control signal 109, generated by controller 103, can cause a modulation device 107 to any one or more of switch on or off, change intensity, volume, brightness and / or other associated parameter(s). Controller 103 is configured to receive sensor data 111 from a set of sensors 113. In an example, the sensor data 111 can represent at least one of a set of environmental data 114 and a set of user data 115. A sensor 116 can comprise a sensor that is provided on a user. A sensor 116 can comprise a sensor that is provided in an environment in which the user is present. A sensor 116 can comprise a sensor that is provided in an environment from which it is desirable to obtain certain data relating to that environment, which may then be used in order to inform a state of operation of a modulation device. The set of sensors 113 therefore form a sensor suite configured to generate, e.g., physiological, environmental and motion data. According to an example, the environment of a user can be inferred from clinical, physiological and motion information collected continuously or at specific timepoints using the set of sensors 113. Some data can be determined from input received by, e.g., a caregiver. For example, clinical data can comprise data representing objective clinical tests such as, e.g., spirography / letter test images, beaker test, and / or subjective clinical tests such as doctor assessment of tremor severity using a validated scale. In an example, a sensor can generate motion data (such as triaxial acceleration / angular velocity data derived from any one or more of a gyroscope, accelerometer, magnetometer). Physiological data can be generated from an integration of multiplexed sensors that measure different body parameters at the same time. For example, a first sensor 1 (Oximetry) can be used to generate data representing any one or more of: Heart rate / heart rate variation, blood pressure, oxygen levels (Peripheral capillary oxygen saturation or SpO2). A second sensor can be used to generate data representing any one or more of, e.g., skin perspiration, skin / body temperature by determining temperature, sweat composition (galvanic skin sensor), level of user hydration. Other examples of sensors are: sensors for muscular and / or neuromuscular activation, electrocardiogram (EMG, MMG), sensors for sleep patterns / circadian rhythms, resting periods, sensors for number of steps, stride, distance, sensors for blood glucose levels, sensors for stroke volume to generate data representing the amount of blood pumped by the left ventricle of the heart in one contraction for example, sensors to detect volumetric changes in peripheral blood circulation (Photoplethysmography (PPG)), sensors for speech recognition, sensors for eye gaze. Figure 2 is a schematic representation of an apparatus according to an example. The apparatus of figure 2 is the same as that of figure 1, with the exceptions that i) one or more of the sensors 113 are provided off-apparatus, and controller 103 is provided off-apparatus. For example, sensors 203 and 205 may be provided remotely from apparatus 101 and communicatively coupled thereto by way of a wired or wireless connection 209. A remote sensor 203 / 205 may be a sensor that is capable of providing environmental data 114 about a environment within which a user using apparatus 101 is situated, for example. Similarly, a remote sensor 203 / 205 may be a sensor that is capable of providing user data 115 about a user using apparatus 101. For example, a sensor can comprise a camera apparatus that is configured to generate image and / or video data representing movements / gait etc. of a user. One or more sensors 201 may be provided on the apparatus 101. Similarly, controller 207 may be provided remotely from apparatus 101 and communicatively coupled thereto by way of a wired or wireless connection 211. Although not shown in figure 2, it will appreciated that similar considerations apply to modulation devices, where one or more may be provided remotely from apparatus 101 and communicatively coupled thereto byway of a wired or wireless connection. In some examples, one or more sensors or sets of sensors can comprise in vivo sensors. For example, a sensor can be embeddable or implantable (fully or partially) under the skin of a user in order to provide a continuous monitoring system (for example to enable continuous glucose monitoring). According to an example, motion and physiological information can be monitored periodically to keep track of all the measurements. Deviations in the form of anomalous data points can be used to determine the presence of or a possible deviation in user homeostasis. Accordingly, data can be generated in order to control one or more devices of an apparatus in order to counteract the cause of the deviation and / or mitigate the effects of the deviation in user homeostasis. In an example, anomalous data points can comprise those that, e.g., fall below a predetermined threshold of a probability distribution, those that are one or more (e.g., 3) standard deviations away from the centroid mass of data points determined using, e.g., Euclidean (ED) or Mahalanobis Distances (MD). According to an example, an anomaly detection model can be trained using a dataset collected under normal (i.e., non-anomalous) conditions in order to obtain a distribution of ED / MD values and define a threshold for anomalous values. The ED / MD can be calculated in the test set and anomalous data points can be detected every time the threshold value(s) are breached. Alternatively, the state of the apparatus can be monitored using one or more autoencoder networks after the normalization and dimensionality reduction steps. A training set of data can be used to calculate the distribution of a reconstruction error under normal (non-anomalous) conditions and to define a threshold for anomalous values. The reconstruction error values can then be estimated in the test set and anomalous data points can be detected every time the threshold values are reached. When the data points cross the ED / MD / reconstruction error thresholds, an alarm can be sent to the user / caregiver / care provider anticipating anomalous data points in the near future. This would enable taking of preventive measures before the anomalous conditions take place. A summary of the average / mean / dispersion / range of every sensor measurement can be provided periodically. This period could be adjusted to the user’s interest and clinical condition (daily, twice a day, every 2 hours etc.). The monitoring periodicity can also depend on the variance of the measurement. For example, some physiological measurements are stable over time while others fluctuate during the day. For every physiological measurement, a fixed time window of data collection (e.g., mean peak / frequency in 15 mins / 1h etc.) can be selected based on the variance of the measurement. Motion and physiological data gathered from sensors can be used to understand or interpret the environment in which the hybrid therapy is to be deployed (e.g., is the user asleep / awake, is the user engaged in physical exercise, is the user in a very hot / cold environment, (e.g. sensed through geolocation via a GPS sensor). Typically, a sensor used to generate physiological data can be provided on or in close proximity to the user in question, such as on or in the body portion for example. A sensor used to generate environmental data can be remote from the user. For example, a sensor may be used to determine the ambient temperature of an environment. Such a sensor can be static and provided at a region of interest, such as a region in which the user often frequents (e.g., a room at home). Continuing with the example of temperature data, a GPS sensor can be provided on the apparatus or as part of another device (such as a smartphone) to which the controller can access (e.g., by way of a Bluetooth connection for example). Accordingly, the apparatus can generate or access data representing the location of the apparatus. In an example, the location of one or more sensors, such as static sensors (e.g., temperature sensors, movement sensors, cameras, PIR devices and so on) can be provided as a part of, e.g., a lookup table. Accordingly, the location of the apparatus can be used to determine the presence of sensors in the vicinity that can be used to generate or for the purpose of gathering, e.g., environmental data. That is, data from one or more sensors can be provided to the controller based on the location of the apparatus. For example, if a user moves into a region in which a temperature sensor is deployed, the apparatus can determine its presence, either by comparing its current position against a repository of locations mapping such locations to available sensors at those locations, and / or by polling for sensors in the vicinity. The apparatus can receive data for a sensor in question. The received data can be used to deploy a contextualized hybrid intervention as described above. For example, a sensor deployed at allocation and whose data is accessible by the apparatus can comprise a camera or movement sensor. Data from the sensor can be used to determine the position and / or speed of movement of a user, which can be used to determine whether the user is, e.g., about to fall, trip, encounter a hazard and so on, thereby enabling the apparatus to deploy a contextualized hybrid intervention in order to stop the user from falling, encountering the hazard etc. Thus, according to an example, the apparatus can sense / identify multiple environmental conditions and can use the gathered / generated data to deploy contextualized hybrid interventions. The environmental conditions / data, in some examples in combination with user data, can be used to recognise user activity, which can lead to appropriate interventions being provided by the apparatus. For example, interventions can be provided to offset, reduce or remove issues experienced by a user in relation to, e.g., activities affected by tremors (drawing, writing, holding objects, limb extension), exercise (running, walking), accidents (falls), physiological state recognition, sleep vs awake, metabolic activity (eating vs not eating), and discerning between diseased (fever) vs non-disease (running) states. Thus, in some embodiments as part of a Machine Learning environment, heuristics can be configured to determine the presence of a signature in, e.g., data generated using an accelerometer that indicates a falling motion. The data can be used to determine the severity of the fall. In some examples, a machine learning model can be used to classify a segment of, e.g., accelerometer data as the signature that indicates falling motion itself. If fall events are scarce in a sample population, a classifier algorithm could be trained on existing fall repositories such as the FARSEEING fall repository for example, where subjects with high risk of falling have been continuously monitored with wearable sensors that include accelerometers, and tested on motion signal data. In an example, a fall can be characterized as a wavelet followed by a period where no motion signal is recorded. The pre-fall, fall and post-fall phases can be summarized with the following features: a lower peak value, followed by an upper peak value, the time period after the upper peak value where no signal is recorded and the standard deviation after the impact. Information collected by the motion / physiological sensors and clinical / user data can be processed and integrated in order to enable deployment of the best or most appropriate hybrid therapy which is adaptive to different environments. Since data can be collected from multiple sensors that can measure different physiological parameters, a multi-sensor data fusion approach can be implemented to integrate all information. In an example, data fusion levels (from lowest to highest abstraction), can depend on the homogeneity of the collected data. For example, raw data from homogeneous sensory sources (e.g., EEG / ECG, accelerometer data from different body parts and so on) can be collected and integrated through a centralized processing approach, where a central node (e.g., the controller 103) can be configured to perform any one or more of de-noising, normalizing, classifying and compressing the data collected for further analysis. At a feature level, each sensor (and / or controller 103) can be configured to extract features from collected / generated raw data, which can then be passed to the controller 103 to generate a multidimensional dataset of feature vectors from all physiological domains. For example, a sensor in the form of an accelerometer, which can be provided on a user, can generate data representing triaxial angular velocity and acceleration data (raw data). Features can be extracted from this raw data, such as maximum / minimum peaks, amplitude, frequency, vector magnitude etc. In an example, controller 103 can comprise Al optimised modules or peripherals and / or hardware acceleration. In some examples, Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) or Autoencoder Networks can be applied to filtered, normalized and decomposed inertial-based sensor data to produce a dataset of non-correlated features. To improve data modelling efficiency, the filtered, normalized, decomposed sensor data with the new extracted features can be subjected to a Feature Selection process to remove irrelevant / redundant features which do not change between observations or which show a strong correlation with other existing features. This selection can be carried out with Genetic Algorithms (GA), Embedded Methods (Random Forest, Lasso Regularization), Wrapper Methods (Forward / Backward / Exhaustive / Recursive feature selection / elimination) or filtering methods using univariate statistics (Information Gain, Chi-Square Test, Fisher Score, Correlation Coefficient, Variance Threshold, Mean Absolute Difference, Dispersion Ratio) for example. In an example, feature selection can be implemented in order to drop or ignore features that are highly correlated and redundant. For example, initially, a feature selection process can be applied to exclude features with a high percentage of missing values (e.g., more than 50%), low variance (e.g., lowest decile) and high correlation (e.g., a correlation coefficient of 0.5 or more). For a particular outcome of interest, a feature importance analysis using Random Forest can be run to rank features based on their relative importance to predict the outcome. If the dataset is highly dimensional (e.g., more than 1000 features), a Dimension Reduction Method (such as, e.g., Principal Components Analysis, Singular Value Decomposition, Latent Discriminant Analysis) can be applied to extract new variables out of the linear combination of existing features, thereby creating a dataset that explains most of the variance. Altogether, the feature selection / extraction steps result in datasets with unique signal features that save data storage space and computational power for further downstream analysis (e.g., resampling, regression / classification / clustering analysis) and help visualize data to identify patterns more clearly. This can therefore result in a ‘most-significant’ feature vector that captures the highest measurement variability with the least amount of features. As such, a ‘most-significant’ feature vector can be used by the controller to create a combined dataset of ‘most-significant’ features across different physiological domains. A binary classifier algorithm can be trained with the combined feature dataset to classify, e.g., a user tremor as ‘stable’ or ‘unstable’. In an example, decision-level fusion can be implemented such that raw data processing, feature selection / extraction and data training / testing take place locally within each sensor. The controller can then weight predictions of each sensor to make a final decision. For example, consider the situation in which a user of the apparatus is having a severe tremor under cold temperatures. Three sensors that are provided, in this example, on or in the body portion of the apparatus (accelerometer, thermometer and ECG) take measurements to generate data representing body motion, temperature and muscle activation respectively. Measurements from each sensor can be processed locally (normalization, feature extraction, feature selection), and, e.g., a binary classifier can be trained and tested with locally sourced data to detect whether the user is experiencing tremors. In an example, this can result in three different ‘verdicts’ coming from each sensor: Accelerometer senses a rise in tremor amplitude and concludes: there is a tremor (1), thermometer senses no increase in temperature because of the cold circumstances and concludes: there is no tremor (0), ECG senses abnormal muscle activation and concludes: there is a tremor (1). Controller 103 ‘collects the votes’ from each sensor and takes a final decision: i.e., in this case, there is a tremor. According to an example, data can be collected in a dynamically adapted time window size based on the variance of the physiological information (some physiological measurements could be stable over time but others may fluctuate within one day as noted above). A balance between shorter and longer time windows should be considered for feature extraction and data storage optimization. For example, shorter time windows allow for faster feature extraction but larger data storage capacity, whereas longer time windows imply longer feature extraction time but less data storage capacity. As physiological, clinical and motion data can vary in both range and distribution, a feature normalization step can be performed to centre the mean and dispersion of every measurement. This ensures that the contribution of each measurement is comparable. In an example, some physiological information measured can be highly correlated (e.g., heart rate-blood oxygen levels). Accordingly, correlation based feature selection can be used to remove features that are highly correlated with each other to avoid redundancies, save data storage and improve computational efficiency in the analysis. According to an example, data from sensors can be processed in different ways. For example, in a centralized data processing implementation, data from all sensors can be collected in controller 103, where normalization and feature selection steps can be performed. In another example, each sensor can independently process its own data and pass the results to the controller where the global analysis can be performed. In another example, hybrid data processing can be implemented in which data collection and some preprocessing steps (e.g., normalization) can be performed with a distributed approach (e.g., by sensors themselves) whereas other steps (e.g., feature selection / extraction) can be performed by the controller. According to an example, a controller, such as controller 103 described above, can be used to receive sensor data and provide the sensor data in a raw or pre-processed form to an external apparatus for further processing. For example, controller 103 can receive data from sensors and transmit the data to a, e.g., cloud-based apparatus configured to process the data and generate the control signals. Such remotely generated control signals can be transmitted back to the controller 103 where they can be used to apply a modulation or modulations. Data provided by controller 103 can be pre-processed. For example, the controller 103 can filter data from sensors (e.g., to remove unwanted or irrelevant components or parts), aggregate data from sensors (e.g., by aggregating data representing outputs from similar sensors), convert data from one format to another and so on. Such pre-processed data can be provided to the remote apparatus for further processing. In an example, the remote cloudbased apparatus can be provided with or have access to additional data from other sources. For example, data relating to a user can be provided to the remote apparatus, or queried by the remote apparatus, from extrinsic systems such as, e.g., existing patient records, mental health applications, physiotherapist input and so on. According to an example, sensor data can be used to determine a set of operational parameters for the set of modulation devices. The set of operational parameters relate to a hybrid stimulatory output of the set of modulation devices, and the set of operational parameters can therefore be used to generate the set of control signals used to enable deployment of, e.g., a user personalized intervention. In this connection, machine learning can be implemented to enable data (user and / or environmental) to be processed in order to determine the set of operational parameters. In some examples, the operational parameters can be tuned and iteratively refined using a self-learning / autonomous control system based on an epsilon greedy reinforcement learning algorithm. For example, such an algorithm can be applied to sensor data using increments in tremor suppression as the reward policy. The level of tremor suppression may be estimated as, e.g., 1 minus the ratio between the tremor power with and without stimulation / mitigation (using a mechanical retardation structure, such as a gyroscope for example), so that values near 1 indicate perfect suppression, values near 0 indicate no change in tremor power and negative values indicate tremor enhancement with respect to the baseline. The average level of tremor suppression per individual can then be estimated under different stimulation conditions to optimize the effectiveness of the device. According to an example, a decreasing value of the tremor suppression index can update the operational parameters and elicit a response from the controller 103 in the form of control signals to trigger a different stimuli pattern of one or more of the modulation devices. Using epsilon greedy as the action selection policy, stimuli patterns that maximize tremor suppression levels can be chosen in each iteration until the levels of tremor suppression get closer to 1, after which the stimuli pattern will remain constant. The stimuli patterns activated by the controller 103 by way of generated control signals for modulation devices in response to updated tremor suppression level values can include any one or more of mechanical / electrical / visual / acoustic stimuli. Besides tremor suppression levels, other reward targets that can be optimized by a reinforcement learning algorithm comprise decreasing reduction ratios in acceleration / angular velocity or other signal features, decreasing signal magnitude vector / areas or stabilization of spectral entropy. In an example, operational parameters and data can be interface-able, e.g., communicable, with external sources (for example, internet / cloud, networked devices, ad-hoc connected devices, other similar devices, etc.). Accordingly, in an example, user / environmental data can be input to a machine learning / AI system in order to generate a set of outputs that provide, represent or map to a set of operational parameters that can be used to generate corresponding control signals for implementation of an intervention using the apparatus. For example, in the case of tremors, tremor severity and trajectories can be predicted using, e.g., binary prediction in which it can be determined whether a user’s tremor is going to worsen in the future (t1) compared to the present (tO), based on information collected in the past (t-1). Binary prediction can be implemented in the form of, e.g., tree-based prediction, such as random forest (aggregate prediction across all classification trees) or gradient boosting algorithms (prediction based on sequential refinement of the trees). Alternatively, a binary classifier can be trained to detect deterioration in tremor severity (no deterioration vs deterioration) or traumatic injuries (fall vs no fall). Similarly, a classifier algorithm can be trained using sensor data to predict tremor severity. This classifier could be a logistic regression model, k-nearest Neighbour algorithm, Decision Tree, Support Vector Machine or Naive Bayes if the outcome is binary (e.g., any tremor vs. no tremor). If the outcome has two or more categories (e.g., Parkinson’s disease vs. Essential Tremor vs no tremor) a multiclass classifier can be applied (k-Nearest Neighbours, Decision Trees, Naive Bayes, Random Forest, Gradient Boosting). After splitting the dataset into training (e.g., 80%) and testing (e.g., 20%) sets, the classifier can be trained using labelled data from participants with known diagnoses. A resampling method such as cross-validation or bootstrap can be applied to the training set to tune the hyperparameters of the algorithm and ensure better prediction performance on the test set. In this connection, data can be trained to calculate a lead time between the detection of an incident (start of tremor deterioration / pre-impact fall) and the actual incident to alert the user in advance. Continuous prediction (forecasting) can be used to predict what user tremor severity value may be in the future compared to the present, based on information collected in the past. For example, ARIMA models (Auto-Regressive Integrated Moving Average) can use cumulative results from past lagged-regressions to forecast future values. In some examples and, optionally, in combination with any example described herein, an exemplary neutral network technique may be applied to the processed sensor data for prediction, classification and clustering purposes of labelled / unlabelled data. The neural network may be one of, without limitation, a feedforward neural network, radial basis function network, recurrent neural network, convolutional network (e.g., Il-net) or other suitable network. Since neural networks are sensitive to feature scaling, the processed sensor data could be normalized and standardized to have, e.g., a mean of 0 and a variance of 1. In some examples and, optionally, in combination of any example described herein, an implementation of a Neural Network may be executed as follows: The processed, normalized and standardized sensor data with m features is fed to the neural network as the input layer, so that each feature is represented by a neuron (xi, X2,....,Xm). The exemplary trained neural network model may specify a neural network by at least a neural network topology (e.g., Multi-layer perceptron (MLP), Recurrent Neural Network (RNN) etc.), series bias values, connection weights and activation functions. In an example, bias values and connection weights can be initialized at random and updated iteratively through gradient descent, stochastic gradient descent or any other suitable algorithm that minimizes the error function. In some instances, a regularization term could be added to the error function to shrink model parameters and prevent overfitting. For example, the topology of a neural network may include a configuration of nodes of the neural network and connections between such nodes. The number of hidden neurons, layers and iterations can be tuned using a validation set. In some instances where the dimensionality of the data is high, an autoencoder could be used in between layers to exclude redundant and highly correlated input features. In some examples and, optionally, in combination of any example described herein, a trained neural network model can also be specified to include other parameters, including but not limited to, bias values / functions and / or aggregation / activation functions, learning rate, number of epochs and batch size. For example, an activation function of a node may be a step function, sine function, continuous or piecewise linear function, rectified linear unit function, exponential linear unit function, softmax function, softsign function, sigmoid function, hyperbolic tangent function, or other type of mathematical function that represents a threshold at which the node is activated. In some examples and, optionally, in combination with any example described herein, an aggregation function may be a mathematical function that combines (e.g., sum, product, division etc.) input signals to the node. In some examples and, optionally, in combination with any example described herein, an output of an aggregation function can be used as input to the activation function. In some examples and, optionally, in combination with any example described herein, a bias can be a constant value or function that can be used by the aggregation function and / or the activation function to make the node more or less likely to be activated. According to an example, a learning rate of 0.01, a number of epochs equal to 150 and a batch size of 10 can be specified a priori and tuned according to the dimension of the dataset, the model performance and the computational time required to run the neural network. It will be appreciated that other values for these parameters can be chosen in order to optimise the output of the model. A neural network as defined above can be trained to solve a classification problem. For example, the neural network can be trained with past sensor data from individuals with, e.g., different tremor severities in order to classify new sensor data into categories of tremor severity. The classifier can be trained using backpropagation to minimize the error function and can support binary (e.g., sigmoid activation function) and multiclass classification (e.g., softmax activation function). For example, in the case of a binary classifier it is desired to discern between participants with tremor disease (labelled as, e.g., ‘1’) and participants without tremor disease (labelled as, e.g., ‘0’), the processed, standardized and normalized sensor data features can be used as the input layer and the sigmoid function as the activation function. After tuning the hyperparameters in a validation set, the output of the neural network can pass through a sigmoid activation function whereby values larger or equal to 0.5 can be rounded up to 1, otherwise to 0. Hence, a sensor data input that results in an output of 1 will be classified as ‘tremor disease’ and a sensor data input that results in an output of 0 will be classified as ‘no tremor disease’. In some examples, where the outcome is continuous and numeric (e.g., tremor severity index, device effectiveness index), a neural network can be trained on processed, standardized and normalized sensor data using backpropagation and with the identity function as an activation function. By defining the mean squared error as the loss function and a set of continuous values as the outcome, the neural network can be trained as a regressor to predict specific continuous outcomes given a set of sensor data. The number of hidden neurons, layers and iterations can be tuned using a validation using resampling methods (e.g., k-fold cross validation). The network topology can be modified until the lowest mean squared error is achieved. In some examples, a multi-output neural network can be designed to derive regression and classification predictions for a single input. As such, the model can take the same number of input neurons as in the previous classification / regression examples, but with two separate output layers that connect to the last hidden layer. The first output layer is the regression output layer with an identity activation function and the second output layer is the classification output layer that uses, e.g., a sigmoid / softmax activation function. Each output layer will have a different loss function: a mean squared error loss for the regression output layer and a sparse categorical cross-entropy for the classification output layer. In an example, and further using tremors as the contextual basis, environmental factors that contribute the most tremor severity can be evaluated. For example, regression models with continuous motion data features as outcomes (highest peak, zero crossing rate, mean amplitude / frequency, peak features (minimums, maximums), dispersion features (standard deviation, variance), magnitude features (magnitude area, vector magnitude mean / median / mode, signal energy, signal magnitude area, phase angle) and environmental factors as exposure can be used. Regression methods can be used to predict numeric outcomes (e.g., amplitude, frequency, or any derived variable to assess tremor severity). Depending on the outcome, different regression models can be built. For example, linear / tree / spline / generalized linear regression model for continuous outcomes that follow linear / non-linear patterns (e.g. amplitude, frequency), logistic regression for binary outcomes (e.g., tremor improvement or tremor stabilization / worsening), multinomial regression for multiclass outcomes of 3 or more categories (e.g., tremor improvement, tremor stabilization, tremor worsening) and Poisson / negative binomial regression for count data (e.g., days of tremor improvement since start of the intervention). In some examples, a random forest implementation can be used to rank environmental factors that contribute the most to tremor severity and other clinical endpoints. Users can be profiled based on physiological / tremor data in different trajectories of tremor severity. For example, in the context of supervised learning, training can be performed using an existing repository of labelled data to identify tremor progression ‘types’. A binary classifier can be used (e.g., stable tremor vs worsening tremor), or a multinomial classifier can be used (e.g., stable during daytime vs stable during night-time vs worsening during daytime vs worsening during night-time, and so on). In some examples, unsupervised algorithms (e.g., Latent Class Analysis, Latent Profile Analysis, Latent Growth Curve Modelling, hierarchical / non-hierarchical clustering or autoencoder neural networks) can be applied as described in to continuous / multinomial outcomes to group individuals into different profiles / trajectories. This can aid identification of individuals at risk of deterioration / poor engagement that could benefit most from preventive interventions and / or personalized nudges. In an unsupervised learning context, profiles of individuals can be identified based on their motion data pattern, clinical and / or environmental data using techniques such as mixture modelling, hierarchical clustering, K-means clustering or C-means clustering. The number of distinct patterns to be identified can be defined a priori after running models with up to, e.g., 8 clusters and checking the statistical performance using, e.g., Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), maximum log- likelihood and entropy values. The model with the highest entropy and maximum log-likelihood values and the lowest AIC and BIC values can be used to define the number of clusters. Since higher number of clusters tend to yield lower AIC / BIC values and higher maximum log-likelihood values, the elbow of the curve of AIC, BIC and maximum log-likelihood values can be used to guide the decision. This elbow indicates the highest leap of model performance. In some examples, autoencoder neural networks could be applied to longitudinal motion signal data to reveal distinct trajectories of tremor severity (e.g., improved / stable / worsened). Depending on the follow-up period (e.g., 3 months, 6 months, 1 year) data can be split in time in, e.g., daily / weekly intervals. A transformation step (e.g., using gaussian process regression) can be used to transform the irregular and sparse observations of each subject into longitudinal probability distributions. These probability distributions can then be used as input vectors for an autoencoder neural network, which can infer distinct probability distributions. A hidden layer is smaller than the input layer to ensure dimensionality reduction and prevent the autoencoder neural network from replicating the input vector. This results in a finite number of learned trajectories. These trajectories can be further used in a supervised classification / clustering exercise to assign individuals to learned trajectories. In some examples, information can be extracted from unstructured and anonymized clinical / healthcare / user data (e.g., medical reports, patient testimonies, diaries, and so on) to (any one or more of) optimize tremor modulation, detect anomalies, classify users in different health behavior categories and perform risk assessment. In this context, Natural Language Processing (NLP) pipelines can be used to automatically summarize a user’s clinical records and / or derive topics / themes from a user’s diary entries. A text lexicon adapted to clinical terms can be implemented, in which a dictionary of medical acronyms / ICD-codes and their standard lexicon equivalent is created (e.g., DM=Diabetes Mellitus). Acronyms can thus be replaced with their standard lexicon equivalent. Pre-processing of raw unstructured data can be implemented such that, e.g., stop words, linkage words, regular expressions and so on are identified and cleaned (noise removal). In the context of feature engineering / extraction, text tokenization / vectorization steps can be implemented so that the cleaned text is broken into manageable units and transformed into numerical vectors for subsequent NLP analysis. In an example, a matrix of features can be created assigning a column for each word / term / phrase, while each row corresponds to a particular user with distinct clinical history, diagnosis, tremor severity etc. For example, the presence of a word / term / phrase and so on can be coded as 1 and its absence as 0. Term frequency / lnverse document frequency can be used to measure how frequent / rare a word is in the text and can identify words that are very frequent / infrequent (e.g., number of occurrences higher or lower than a predetermined threshold value. The frequency of terms indicating health risks (e.g., “unwell”, “uncomfortable”, “worried”, “difficulty”) can be used to flag users at risk of deterioration. Similarly, an annotated dataset with such risk labels can be trained with classifiers (e.g., decision trees, random forests, logistic regression, random forest or recurrent / convolutional neural networks) to prospectively identify users at low, medium and high risk of deterioration. This stratification of users can help deliver a tailored, timely and corrective hybrid modulation that prevents deterioration. In some examples, a classifier model can be trained with existing clinical records / diaries to classify new unstructured data streams for adverse event detection and medication management (e.g. adverse event vs non-adverse event, dosage adjustment vs medication change vs no medication change). In the context of sentiment analysis, a text-adapted sentiment lexicon could be deployed to assign sentiment scores to each word token. The aggregated sentiment scores for a particular user can then be used to assign an overall sentiment score for each day. Sentiment score thresholds can be established to stratify users into categories (e.g., positive, neutral, negative, ambiguous, indifferent, apprehensive). These categories can then be used to deliver tailored messages that nudge the user into proactive health behaviors. In some examples, user physiological and personal data can be funnelled through generative Al models to elicit a customized audio / text / multimedia message for the user and other stakeholders (family, doctor, caregiver etc.) that could help improve tremor management and prevent incidents. This customized message could comprise a periodic summary of the physiological data (e.g. daily / weekly / monthly average values, trends), a reminder (e.g. take medication, use the device) or a recommendation that could improve health outcomes and / or prevent further deterioration (e.g. change device hybrid stimulation settings, increase / decrease stimulation level etc.). For example, a distinct tremor pattern signature over time could be indicative of tremor deterioration in the short term. Upon recognition of such tremor pattern signature, the controller can elicit a generative Al message via text / audio channels to warn the user about such deterioration. Such data could be presented to the patient over a smartphone application, computer application, electronic website, smartwatch and other wearable applications, et cetera. Machine learning implementations can be trained periodically (every day / week / month) as more data becomes available, and updates can take place in off periods when the apparatus is not being used. In an example, an update can be triggered automatically every time there is no signal recorded or via direct user input. To avoid lags in hybrid therapy deployment, the apparatus can progressively extend the period of training / testing. For example, in a first week of use, training / testing datasets can be generated daily and parameters updated accordingly. In subsequent weeks, training / testing datasets can be generated every few days (e.g., every three days) and parameters updated accordingly. Thereafter, the apparatus can be considered familiarised with the physiological / motion variances of the user, so frequent updates may not be needed. The apparatus can therefore generate training / testing sets and parameters can be updated accordingly. These training / testing datasets could be further consolidated in a biomarker database of physiological and pathophysiological disruptions. Such data could be used to generate relevant physiological and pathophysiological data including simulated tremor patterns and simulated patients with various neurological conditions. These periodically updated biomarker databases can be exported for relevant applications including clinical research, clinical testing, device optimization, and research and development purposes. The preceding description has been provided to enable others skilled in the art to best utilize various aspects of the exemplary embodiments disclosed 5 herein. This exemplary description is not intended to be exhaustive or to be limited to any precise form disclosed. Many modifications and variations are possible without departing from the spirit and scope of the instant disclosure. The embodiments disclosed herein should be considered in all respects illustrative and not restrictive. Reference should be made to the appended claims and their 10 equivalents in determining the scope of the instant disclosure.

Claims

1. Apparatus, comprising:a body portion mountable to a body part of a user;a controller; anda set of modulation devices provided on the body portion, each modulation device of the set of modulation devices configured to receive a respective control signal from the controller to regulate a function thereof, wherein the controller is configured to:receive sensor data from a set of sensors, the data representing at least one of a set of environmental data and a set of user data;using the sensor data, determine a set of operational parameters for the set of modulation devices, the set of operational parameters relating to a hybrid stimulatory output of the set of modulation devices; andusing the set of operational parameters, generate the set of control signals.

2. The apparatus as claimed in claim 1, wherein at least one of the set of modulation devices comprises an electrical neuromuscular stimulator.

3. The apparatus as claimed in claim 1 or 2, wherein at least one of the set of modulation devices comprises an electrical heating element.

4. The apparatus as claimed in any preceding claim, wherein at least one of the set of modulation devices comprises a Peltier device.

5. The apparatus as claimed in any preceding claim, wherein at least one of the set of modulation devices comprises a gyroscopic device.

6. The apparatus as claimed in any preceding claim, wherein at least one of the set of modulation devices comprises a device configured to deliver at least one of a visual, or an acoustic signal.

7. The apparatus as claimed in any preceding claim, wherein at least one of the set of modulation devices comprises a device configured to deliver a mechanical stimulation.

8. The apparatus as claimed in any preceding claims, wherein the set of modulation devices are configured to deliver sensory stimulation for the user.

9. The apparatus as claimed in any preceding claim, wherein at least one of the set of sensors comprises an inertial measurement unit or a force sensor.

10. The apparatus as claimed in any preceding claim, wherein at least one of the set of sensors is user mountable to the body portion.

11. The apparatus as claimed in claim 10, wherein the at least one user mountable sensor is implantable or embeddable in the body portion.

12. The apparatus as claimed in claim 10 or 11, wherein the at least one user mountable sensor is configured to generate the set of user data representing one or more characteristics or vital signs of the user.

13. The apparatus as claimed in any preceding claim, wherein at least one of the set of sensors is configured to generate the set of environmental data representing a state of at least one environmental parameter.

14. The apparatus as claimed in any preceding claim, further comprising a transceiver configured to receive at least one of the set of environmental data and the set of user data.

15. The apparatus as claimed in any preceding claim, wherein the set of modulation devices are configured to apply the hybrid stimulatory output using the set of control signals.

16. A machine-readable storage medium encoded with instructions for controlling a set of modulation devices of an apparatus, the instructions executable by a processor of the apparatus, whereby to cause the apparatus to:receive sensor data from a set of sensors, the data representing at least one of a set of environmental data and a set of user data;using the sensor data, determine a set of operational parameters for the set of modulation devices, the set of operational parameters relating to a hybrid stimulatory output of the set of modulation devices; andusing the set of operational parameters, generate the set of control signals to control respective functions of the modulation devices to generate the hybrid stimulatory output.

17. The machine-readable storage medium as claimed in claim 16, further comprising instructions to cause the apparatus to:using the sensor data, determine a set of parameters representing a current state space for the user;determine a set of actions on the basis of the current state space for the user;using a policy function, determining the set of control signals, wherein the set of control signals are configured to cause the hybrid stimulatory output of the set of modulation devices to be executed, thereby affecting the set of actions and thereby modifying the state space of the user.

18. The machine-readable storage medium as claimed in claim 16 or 17, further comprising instructions to cause the apparatus to:regulate an intensity of the respective functions of the modulation devices.

19. The machine-readable storage medium as claimed in any of claims 16 to 18, further comprising instructions to cause the apparatus to:regulate an operational state of one or more of the modulation devices.

20. The machine-readable storage medium as claimed in any of claims 16 to 19, further comprising instructions to cause the apparatus to:generate a set of operational parameters on the basis of a position of at least one of the modulation devices.

21. The machine-readable storage medium as claimed in any of claims 16 to 20, further comprising instructions to cause the apparatus to:modify a predetermined hybrid stimulatory output on the basis of the sensor data to generate a modified hybrid stimulatory output.

22. The machine-readable storage medium as claimed in claim 21, further comprising instructions to cause the apparatus to:generate a set of control signals for the modified hybrid stimulatoryoutput.

Images