Navigation guidance devices

Abstract

A navigation guidance device 100 with input componentry configured to obtain path data representing a path 172 along which a path-following person 102 is to walk. The path data represents a digital pa

Description

The following terms are registered trade ma: an should be rea assuc w they occur in this document; Bluetooth ARKit Apple ARCore Google NAVIGATION GUIDANCE DEVICES Field The present disclosure relates to navigation guidance devices. Background Computer vision can identify objects. Simultaneous Localisation and Mapping (SLAM) can map and locate a person within an unknown environment. Both technologies can, in principle, benefit people, such as blind people. However, existing devices do not adequately communicate useful information derived using these technologies to vision impaired people. The reader is referred, in this regard, to WO 2016 / 116182 Al and JP 2010057593 A. Existing navigation guidance devices are not adequate for providing navigation guidance to vision impaired people. For example, they do not adequately communicate useful navigation guidance information, they provide only high-level navigation guidance using low-resolution maps, they require an environment to be scanned before a vision impaired person can navigate through the environment, they require changes to existing infrastructure (for example, installing Bluetooth™ beacons), and / or they potentially communicate dangerous navigation guidance. Summary According to first embodiments, there is provided a navigation guidance device comprising: input componentry configured to obtain path data representing a path along which a path- following person is to walk, the path data representing a digital path marker, the digital path marker being based on tracked movement of a path-defining person along a walked path, the walked path having been walked along by the path-defining person, wherein a digital path marker is a digital representation of a position in 3D space; and output componentry comprising a kinaesthetic output mechanism operable to provide kinaesthetic output to guide the path-following person along at least part of the path using the digital path marker. According to second embodiments, there is provided a navigation guidance device configured to obtain path data representing a path, the path data representing a digital path marker, the digital path marker being based on tracked movement of a path-defining person along a walked path, the walked path having been walked along by the path-defining person, wherein a digital path marker is a digital representation of a position in 3D space. According to third embodiments, there is provided a navigation guidance device configured to use a real-time kinematic positioning satellite navigation, RTK-GNSS, receiver to provide navigation guidance to a user of the navigation guidance device, wherein the navigation guidance device comprises a handheld navigation guidance device and / or a wearable navigation guidance device. According to fourth embodiments, there is provided a navigation guidance device configured to use visual simultaneous localization and mapping, VSLAM, to provide navigation guidance to a user of the navigation guidance device, wherein the navigation guidance device comprises a handheld navigation guidance device and / or a wearable navigation guidance device. According to fifth embodiments, there is provided a navigation guidance device configured to use visual odometry, VO, to provide navigation guidance to a user of the navigation guidance device, wherein the navigation guidance device comprises a handheld navigation guidance device and / or a wearable navigation guidance device. Brief Description of the Drawings Various embodiments will now be described, by way of example only, with reference to the accompanying drawings in which: Figure 1 shows a perspective view an example of a sensory substitution device in three different configurations; Figure 2 shows a plan view of the example sensory substitution device shown in Figure 1; Figure 3 shows a perspective view of another example of a sensory substitution device; Figure 4 shows an exploded perspective view of the example sensory substitution device shown in Figure 3; Figure 5 shows schematically an example of out-of-plane shape-changing motion; Figure 6 shows a schematic representation of example items associated with an example sensory substitution device; Figure 7 shows a perspective view of another example of a sensory substitution device; Figure 8 shows a perspective view of another example of a sensory substitution device in three different configurations, the example sensory substitution device comprising a tilt compensator; Figure 9 shows a perspective view of another example of a sensory substitution device in five different configurations; Figure 10 shows a perspective view of another example of a sensory substitution device; Figure 11 shows a schematic representation of an example of an environment in which another example of a sensory substitution device is being used; Figure 12 shows a schematic representation of another example of an environment in which another example of a sensory substitution device is being used; Figure 13 shows several schematic representations of another example of a sensory substitution device being held in different ways; Figure 14 shows a perspective view of another example of a sensory substitution device attached to a cane; Figure 15 shows a schematic representation of a system comprising another example of a sensory substitution device; Figure 16 shows a perspective view of another example of a sensory substitution device; Figures 17A to 17D show a schematic representation of example of an environment in which a path is defined; Figures 18A to 18C show a schematic representation of example of an environment in which a path is followed; Figures 19A to 19D show a schematic representation of example of an environment in which a path is followed and adjusted; Figure 20 shows a schematic representation of an example of a system in which a path is defined and / or followed; Figures 21A and 21B show a message flow diagram depicting an example of a method of defining and following a path; Figure 22 shows a schematic representation of another example of an environment in which a path is defined; Figure 23 shows a schematic representation of another example of an environment in which a path is defined; Figure 24 shows a schematic representation of another example of an environment in which a path is followed and adjusted; Figure 25 shows a schematic representation of another example of an environment in which a path is followed and adjusted; Figure 26 shows a schematic representation of an example of an environment in which a digital contextual marker is activated; Figure 27 shows a schematic representation of an example of an environment in which a digital directional marker is used; Figure 28 shows a schematic representation of an example of an environment in which a path is adjusted; Figure 29 shows a schematic representation of an example of an environment in which a multiple paths may be followed; Figure 30 shows a schematic representation of an example a system in which path data is used; and Figure 31 shows a schematic representation of an example arrangement of route data. Detailed Description As explained above, existing devices do not adequately communicate useful information derived using technologies such as computer vision and SLAM to vision impaired people. In particular, such existing devices do not adequately convey such information non-visually. Some assistive technology products can identify objects using computer vision and can provide low-resolution alerts about obstacles. Such products include smart canes, smart glasses and smartphone applications which use audio cues and vibration. Audio cues allow users to find locations. However, this form of feedback is ambiguous and can obscure important environmental sounds and social interactions. Vibratory sensations are cognitively distracting, can lead to short-term numbness, and have a low information bandwidth. While vibration may be effective for providing alerts, it is not as effective for conveying complex information such as direction or distance. Without significant advancements in the human-machine interface, smartphone applications, smart canes and smart glasses are not effective enough to approach the functionality of a guide dog. Shape-changing, haptic interfaces for visual impairment are known. Known interfaces can change shape in the hand of a user to communicate information, such as which heading the user should walk on. Such known interfaces use Bluetooth™ beacons and Global Positioning System (GPS), rather than computer vision or SLAM, to accomplish this. Such interfaces also have limited shape- changing capabilities. For example, they may have a part that can rotate relative to another part and slide forward and backward, whilst constrained to move in a single 2D plane, here called “plane- constrained”. The inability of the mechanism to move out-of-plane, in other words out of that 2D plane and freely in the 3D space around the interface, restricts not only the amount of information that can be conveyed to the user but also the intuitiveness of use of such interfaces. Various other drawbacks to such existing interfaces will now be described. GPS is generally only accurate to +5 metres. This is not accurate enough to replace a guide dog. For example, a user cannot be directed to stay on a pavement. GPS also relies on receiving GPS signals from GPS satellites. This is not possible in some environments, such as environments that are deep underground. Bluetooth™ beacons need to be installed in the locations where the existing interfaces are being used. This may involve changes to that environment and may mean that such interfaces can only be used in locations in which Bluetooth™ beacons have already been installed. Neither Bluetooth™ beacons nor GPS can be used to detect and navigate around temporary obstacles, such as dustbins. Such existing interfaces are not optimised to direct users to point at objects in 3D space. Instead, they are optimised for directing a walking heading. Such existing interfaces use geared mechanisms and screw-driven actuators to actuate shape changes. As a result, such interfaces are relatively large and heavy, and have relatively low responsiveness. The motion of such existing interfaces is not particularly organic and is not designed to be arrested by the grip of the user, in other words held tightly by the user. A user cannot use such existing interfaces to point at and identify objects. One existing interface is in the form of a candlestick-type device, which is elongate and held vertically. In contrast, example devices described herein may be designed to be held horizontally. Such devices are more intuitive for pointing. Additionally, example devices described herein may primarily communicate angles, for example for communicating pointing and direction. This differs from communicating translational movement only, for example movement through space. However, example devices described herein may nevertheless incorporate a modality that enables distance information to be delivered to a user, for example using vibration, sound, other haptic features, etc. Further, example devices can comprise shape-change functionality at the front of the device such that a thumb and forefinger of a user can detect shape-change, while the bulk of the device can be held in the palm of the user. With the thumb and forefinger being particularly sensitive, enhanced proprioceptive output may be provided, as will be described in more detail below. Further, some example devices described herein comprise and / or leverage one or more forward-facing spatial sensors. Such sensors have a viewing perspective indicative of the pointing direction of the user. This contrasts to previous designs, such as the candlestick-type device, where there are no such spatial sensor locations. Space constraints and difficulties in mechanical design are typically the major limiting factor in design of state of the art shape-changing interfaces. Spatial sensors, such as optical and radio wave sensors, are constrained in their potential placement locations on the device and are significantly larger than previously used sensors, such as GPS and inertial measurement units (IMUs). This contravenes the design trends in the state of the art by increasing the mechanical and / or space constraints posed when designing haptic interfaces, such as handheld haptic interfaces. Various terms will be used in conjunction with examples described below. Such terms will now be explained. The term “environment” encompasses different types of environments, such as real-world and digital environments. For example, example devices described herein may be used as game controllers. In some examples, a device in accordance with examples is located in a real-world environment. The device may use spatial sensor data indicative of a spatial representation of the (real- world) environment in which the sensory substitution device is (physically) located. The device may, for example, orient itself to point to an object in that real-world environment. In some examples, a device in accordance with examples is located in a digital (or “virtual”) environment. The device may use spatial sensor data to cause the device to change shape to provide kinaesthetic output indicative of a target position. The target position may be a digital target position, not detected by the spatial sensor. The device may, for example, orient itself to point to a digital object in that digital environment. The device may also be able to find digital waypoints, and the like, which are not directly detected by the spatial sensor. The term “augmented reality” (AR) is used herein to relate generally to using digital content in real-world experiences, such as guiding a user to virtual target placed relative to real-world space. The term “computer vision” is used herein to relate generally to identifying one or more objects or people from camera footage, or to make inferences about an environment such as identifying pavement versus road, or grass versus tarmac. As such, computer vision can encompass "classification". Classification may be either of an entire frame or of a subregion of a frame (also called "semantic segmentation"). Computer vision also includes functionality such as, but not limited to, 3D pose estimation, motion estimation and scene modelling. These types of computer vision do not return a heuristic understanding of a scene, such as an object being a road. However, they still enable significant functionality, such as teleassistance and tracking of pre-set markers. Both of these use cases can, conceivably, be achieved using computer vision without any heuristic identification that may usually be associated with computer vision. Computer vision may also encompass visual odometery. Visual odometry is especially useful for 3D pose estimation, namely finding position and orientation in 3D space, which is especially effective for SLAM. The term “LIDAR”, which stands for light detection and ranging, is used herein in relation to technology used to measure distances and build a point-cloud of an environment. The term “point cloud” is used herein to mean a series of depth measurements derived from a LIDAR, ultrasonic, RADAR, SONAR or other spatial sensor scan that can be used to build a 3D computer model of an environment. The term “haptic” is used herein, in relation to a device, to mean a device which communicates information through the touch sense. The term “digital marker” is used herein to refer to a point in space which has been digitally recorded or generated. Examples include, but are not limited to GPS locations, AR anchors in software, or a location relative to a device’s position. The term “AR Anchor” is used herein in relation to digital markers placed within specific software suites, such as ARkit™ by Apple™ and ARCore™ by Google™. The term “tilt” is used herein to refer to roll of a device about its length, such as when a user rolls their wrist. Tilt and tilt compensation will be described below in more detail with reference to Figure 8. The term “vector” is used herein to refer to a quantity with both magnitude and direction. The term “pointing vector” is used herein to mean the unit vector along which the device is currently pointing. The term “target vector” is used herein to mean the vector between the device and the target position. The term “error” is used herein to mean the angle between the pointing vector and the target vector. The terms “proprioception”, “kinaesthesia” and “kinesthesia” are used herein interchangeably to mean the sense of self-movement and body position. The terms “modality” and “sensory modality” are used herein interchangeably to mean a type of stimulus or what is perceived after such a stimulus. Examples of sensory modalities include, but are not limited to, smell, taste, touch, vision and hearing. The term “sensory substitution” is used herein to mean passing information that is normally gathered by one sensory modality (such as vision) through the modality of another (such as touch). The terms “shape change”, “change shape”, “change of shape” and the like are used herein to mean a change to at least one property of a shape. A shape change may, for example, result in a change in at least one dimension of the shape, with the geometric shape remaining the same. For example, a change of shape may encompass a cylindrical shape changing from having a first length to a second length. A shape change may, for example, result in a change of geometric shape. The terms “device” and “interface” are used interchangeably. The term “spatial sensor” is used herein to mean any type of sensor that can be used to sense a spatial environment, such as spatial information of such an environment. For example, LIDAR, RADAR and optical cameras primarily return information about the space towards which they are pointed; not the body on which they are mounted. IMUs and GPS modules, on the other hand, primarily return information about the body to which they are attached, not information about the surrounding space, and so are not spatial sensors. The term “sensor” is used herein more generally to mean any type of sensor. The term “spatial sensor data” is used herein to mean any type of data provided or output by a spatial sensor. The term “sensor data” is used herein more generally to mean any type of data provided or output by any type of sensor. The term “target position” is used herein to mean any position in 3D space that is a target for a sensory substitution device, a user of a sensory substitution device, or another device or person to locate. An example of a target position is a digital marker placed on an object or an intermediary point along a path. The target position is not a position on the sensory substitution device itself. In some examples, the sensory substitution device is configured to point towards the target position. The term “forward-facing” is used herein to mean facing forwards with respect to the front of the device. This may correspond to facing in a direction of a target position in some situations, but the target position may be behind the user in other situations. The term “SLAM”, which stands for Simultaneous Localisation and Mapping, is used herein to relate to the process of using spatial sensor data to construct or update a map of an environment, whilst simultaneously keeping track of the location of a user within the environment. This can also be used to help keep track of targets set by a user and / or by another entity. The term “plane-constrained” is used herein, in connection with a device, to mean freedom of movement which is constrained to a single 2D plane that has its frame of reference on the device. For example, plane-constrained motion may allow left and right movement, forwards and backwards movement, and rotation whilst remaining constrained to one 2D plane, but not movement upwards and downwards as this would move out of the 2D plane and into 3D space. Motion that is not plane- constrained may move out of a 2D plane and into 3D space. Such motion may be referred to as “out- of-plane” motion. The term “shape-change mechanism” is used herein to mean a mechanism that changes a shape. The shape-change mechanism may comprise a “shape-changing mechanism” which changes its own shape and thereby the shape of a device in which it is incorporated. However, a shape-change mechanism may change the shape of a device in which it is incorporated without changing its own shape. The term "device axis" or “device centre line” is used herein to refer to a line approximately along the longitudinal axis of the device, representing the home position or straight position of the device. Examples described herein provide navigation technology in the form of a device that changes shape. Humans have a natural ability to interpret object shapes with their hands, such as whether an elongate object is curved to the right or to the left, for example. This ability is exploited to communicate navigation information non-visually. Such information is communicated via the touch sense and can be used to indicate 3D locations. The user can follow the direction of curvature thereby directing the user to point in a desired direction. Such examples exploit the combination of the ability of humans (i) to sense the location of their extremities non-visually (proprioception) and (ii) to infer 3D locations by pointing with their hands. Pointing with the hands is a natural human gesture which is observed even in congenitally blind children who cannot observe and learn this behaviour from other humans. It is therefore a surprisingly effective way of communicating vectorised information to vision impaired persons. Directing the user to point in a desired direction may correspond to directing the user to orient a longitudinal axis of the device towards a target position. Examples described herein also provide a handheld sensory substitution device which uses a shape-changing, haptic interface in combination with computer vision and / or LIDAR to direct a user towards one or more locations and / or one or more objects of interest by prompting the user to point their hand, and hence the device, in the direction of the object(s) and / or location(s) of interest. Examples may, however, still use only GPS and / or only an IMU for controlling the device. Examples also provide a new human-machine interface that can direct users towards objects and / or locations of interest by passing information through the sense of touch via shape changes. In examples, the interface is in the form of a sensory substitution device. The sensory substitution device may bend in two or more dimensions whilst held in the hand, aiming or pointing in the desired direction. The user can sense which direction the device is pointed in via the shape change and can intuitively follow this guidance. This non-visually conveys direction to users through the touch sense with minimal prior knowledge and minimal training. Prompting a user to point in the desired direction works with their innate non-visual spatial interpretation capabilities. Examples, therefore, reduce an intractable problem of describing 3D navigation with audio cues or vibration sensations into an intuitive and straightforward experience for vision impaired people. Examples described herein also provide a handheld device with forward-facing LIDAR, which scans the environment in which the device is located to build a local 3D computer model in the form of a point cloud. The point cloud is used in isolation, or is combined with computer vision using SLAM to construct a map of the local environment the user is walking through while simultaneously keeping track of the location of the user within that environment. Relevant navigation information is then passed to the user through their touch sense using an advanced, shape-changing haptic interface. This augments the capabilities of GPS to provide enhanced and intuitive local guidance, for example keeping a person on a pavement, directing a person to a platform, or dealing with temporary obstacles such as other people or dustbins. This can be provided without modifications to infrastructure. The computer vision capability may, for example, identify bus stops, bus numbers and train doors. Computations, such as SLAM computations, may be performed on a connected device, such as a smartphone, to keep costs of the interface low and accessibility high. Referring to Figures 1 and 2, there is shown an example of a sensory substitution device (also referred to herein as a “device” for convenience and brevity) in three different configurations. In Figure 1, the device is indicated with reference signs 100a, 100b, 100c in the first, second and third configurations respectively. In use, the device 100a, 100b, 100c is held and gripped by a user. In Figure 1, the user is indicated with reference signs 102a, 102b, 102c when the device is in the first, second and third configurations respectively. In use, the device 100a, 100b, 100c is located in an environment. The environment comprises a target position 104. The target position 104 is, in this example, a position in 3D space, which is being targeted by the user 102a, 102b, 102¢ and / or another entity. The target position 104 may correspond to a target object and / or a target location. In the first configuration, the device 100a is bent to the right and points at the target position 104. In the second configuration, the device 100b is straight and points at the target position 104. A straight device shape is easily sensed by the hand of the user 102b which provides an unambiguous indication that the user 102b is pointing directly at the target position 104. In the third configuration, the device 100c is bent to the left and points at the target position 104. In this example, the device 100a, 100b, 100c is a handheld device. This provides a highly portable device, which can readily leverage proprioception. In use, the user 102a, 102b, 102c holds the device in their hand 100a, 100b, 100c like a flashlight. The user 102a, 102b, 102c can point the device 100a, 100b, 100c¢ in any direction. In this example, the device 100a, 100b, 100c comprises a rear portion. In Figure 1, the rear portion is indicated with reference signs 106a, 106b, 106¢ in the first, second and third configurations respectively. In this example, the device 100a, 100b, 100c comprises an exposed shape-change mechanism portion. In Figure 1, the exposed shape-change mechanism portion is indicated with reference signs 108a, 108b, 108c in the first, second and third configurations. The exposed shape-change mechanism portion 108a, 108b, 108c is exposed in that it is at the surface of the device 100a, 100b, 100c. As will be explained in more detail below, other elements of the shape-change mechanism are within the device 100a, 100b, 100c. In this example, the device 100a, 100b, 100c comprises a gripping portion. In Figure 1, the gripping portion is indicated with reference signs 110a, 110b, 110c in the first, second and third configurations respectively. In this example, the gripping portion comprises multiple gripping regions, with a first one of the gripping regions being on one side of the exposed shape-change mechanism portion 108a, 108b, 108c and the other one of the gripping regions being on the other side of the exposed shape-change mechanism portion 108a, 108b, 108c. The gripping portion 110a, 110b, 110c is designed and configured to be gripped by the user 102a, 102b, 102c. This results in the device 100a, 100b, 100c being convenient to use. In addition, hands are the best-suited part of the body to receive and interpret the shape-changing output of the device 100a, 100b, 100c. In this example, the gripping portion 110a, 110b, 110c extends most of the length of the device 100a, 100b, 100c. The exposed shape-change mechanism portion 108a, 108b, 108¢c may form part of the gripping portion 110a, 110b, 110c if it is intended to be gripped by the user 102a, 102b, 102c. In this example, the device 100a, 100b, 100c comprises a front portion. In Figure 1, the front portion is indicated with reference signs 112a, 112b, 112c in the first, second and third configurations respectively. The rear portion 106a, 106b, 106c, the exposed shape-change mechanism portion 108a, 108b, 108c, the gripping portion 110a, 110b, 110c and the front portion 112a, 112b, 112c collectively define a body of the device 100a, 100b, 100c. The user 102a, 102b, 102c can direct the device 100a, 100b, 100c to find one or more target objects and / or locations 104. Such object(s) 104 may be common (also referred to as “everyday”) objects, environmental features such as pavements and roads, or otherwise. The device 100a, 100b, 100c may comprise a microphone. Alternatively or additionally, one or more other devices to be used with the device 100a, 100b, 100c may comprise a microphone. In examples, the user 102a, 102b, 102¢ can control the device 100a, 100b, 100c¢ using only their voice, for example though natural language processing. In examples, the user 102a, 102b, 102c can therefore verbally describe their intentions to a controller (comprised in the device 102a, 102b, 102¢c or another device) using natural language processing. For example, the user 102a, 102b, 102c could ask the device 100a, 100b, 100c to find them a door, or take them to a particular train platform. This may, for example, help the user 102a, 102b, 102c navigate through public transport networks, without modifications to network infrastructure. As such, in some examples, the target position 104 is identified based on processing a voice input using natural language processing. This provides an especially effective input modality for vision impaired people. The user 102a, 102b, 102c then points the device 100a, 100b, 100c in directions where the target object and / or location 104 is suspected to be positioned. In this example, the device 100a, 100b, 100c comprises a spatial sensor. In Figure 1, the spatial sensor is indicated with reference signs 114a, 114b, 114c when the device is in the first, second and third configurations respectively. The spatial sensor 114a, 114b, 114c is, in this example, configured to provide spatial sensor data indicative of a spatial representation of an environment in which the device 100a, 100b, 100c is located. In this example, the spatial sensor 114a, 114b, 114c is in the front portion 112a, 112b, 112¢ of the device 100a, 100b, 100c. Such a spatial sensor 114a, 114b, 114c may be referred to as “forward- facing”. As such, the spatial sensor 114a, 114b, 114c may be placed on the front of the device 100a, 100b, 100c. In such examples, the spatial sensor 114a, 114b, 114c has a viewing perspective indicative of the pointing direction of the user 102a, 102b, 102c. By the spatial sensor 114a, 114b, 114c having the same perspective as the pointing direction, the device 100a, 100b, 100c can readily and easily orient itself such that the target position 104 is in the middle of a field of view of the spatial sensor 114a, 114b, 114c and, hence, may be in the middle of a camera frame when the user 102a, 102b, 102¢ points their hand at the target position 104. In this example, the spatial sensor 114a, 114b, 114c comprises a camera. However, other types of spatial sensor may be used. Examples of other types of spatial sensor include, but are not limited to, LIDAR, ultrasonic and infrared sensors. As such, in some examples, the target position 104 is determined using computer vision and / or LIDAR. This enables the device 100a, 100b, 100c to be used to identify and locate objects, to perform SLAM, and to map environments. However, ultrasound can also be used to map environments, and infrared enables the device 100a, 100b, 100c to be used in the dark. The latter may be effective whether the device 100a, 100b, 100c is used by a vision impaired person or by a person who is not vision impaired. The spatial sensor 114a, 114b, 114c may have an ultrawide (or “ultra-wide”) field of view, where the sensor 114a, 114b, 114c has an angle of view of greater than 100 degrees. This facilitates objects remaining within the field of view of the spatial sensor 114a, 114b, 114c despite device movement. Where the device 100a, 100b, 100c comprises multiple spatial sensors 114a, 114b, 114c, one or more of the spatial sensors 114a, 114b, 114c may have an ultrawide field of view and one or more others of the spatial sensors 114a, 114b, 114c may not have an ultrawide field of view. This also enables a target vector, between the spatial sensor 114a, 114b, 114c and the target position 104, to be kept within the line of sight of the spatial sensor 114a, 114b, 114c even when the device 100a, 100b, 100c and / or the user 102a, 102b, 102¢c moves. The device 100a, 100b, 100c may comprise one or more light sources. This may enhance use of the device 100a, 100b, 100c in dark environments. In particular, computer vision efficiency is heavily dependent on lighting conditions. The device 100a, 100b, 100c may comprise one or more infrared light sources. This may be especially effective where the device 100a, 100b, 100c comprises one or more infrared cameras. Through computer vision, the camera 114a, 114b, 114c ‘looks’ for the object(s) of interest 104. In this example, the device 100a, 100b, 100c and / or another device analyses the camera frame (in other words, an image captured by the camera 114a, 114b, 114c). The camera 114a, 114b, 114c can therefore capture data that can be interpreted to identify objects 104 using computer vision. In this example, the device 100a, 100b, 100c comprises object identification functionality. However, object identification functionality may be provided in another device to be used with the device 100a, 100b, 100c. More specifically, in this example, the device 100a, 100b, 100c is operable to identify objects when the device 100a, 100b, 100c is pointed at them. It can be more convenient and natural for the user 102a, 102b, 102c to point, compared with using a device such as a smartphone, to identify such objects. Using computer vision in this manner enables the user 102a, 102b, 102c to be located in 3D space within an unknown local environment and enables the user 102a, 102b, 102c to navigate through the unknown environment using SLAM. Specifically, in this example, the device 100a, 100b, 100c is configured to be controlled using SLAM. This enables the device 100a, 100b, 100c to provide spatial guidance in any environment. Identified objects 104 may also be tracked. The device 100a, 100b, 100c changes shape via a shape-change mechanism. The shape- change mechanism is operable to cause the device 100a, 100b, 100c to change shape, based on the spatial sensor data provided by the camera 114a, 114b, 114c (and / or another type of spatial sensor), to provide kinaesthetic output to the user 102a, 102b, 102c of the device 100a, 100b, 100c. The kinaesthetic output is indicative of a target position 104 in the environment. As such, the device 100a, 100b, 100c indicates to the user 102a, 102b, 102c where the object 104 is in 3D space. More specifically, the device 100a, 100b, 100c indicates to the user 102a, 102b, 102c where the object 104 is in 3D space relative to the user 102a, 102b, 102c. If the object 104 is identified within the camera frame and if the device 100a, 100b, 100c is not already pointing at the object 104, the device 100a, 100b, 100c changes shape in the hand of the user 102a, 102b, 102¢ to point at the object 104. The user 102a, 102b, 102c can feel the change in shape of the device 100a, 100b, 100c as a result of them gripping and holding the device 100a, 100b, 100c. More specifically, in this example, the user 102a, 102b, 102c can feel the change of shape of the device 100a, 100b, 100c in the gripping portion 110a, 110b, 110c of the device 100a, 100b, 100c. After feeling which direction the shape of the device 100a, 100b, 100c is pointed towards, the user 102a, 102b, 102c can move their hand in this direction. When moving their hand, and hence the device 100a, 100b, 100c, the user 102a, 102b, 102c is receiving kinaesthetic output. This kinaesthetic output is used to infer direction. The change of shape may be in at least one degree of freedom. This may enable the device 100a, 100b, 100c to indicate depth or distance. The change of shape may be in at least two degrees of freedom. This may enable the device 100a, 100b, 100c to point in any direction and, therefore, indicate direction. The change of shape may be in three degrees of freedom. This may enable the device 100a, 100b, 100c to point in any direction, as well as indicate depth, distance or heading. In this example, the change of shape comprises bending of the device 100a, 100b, 100c. This provides a natural-feeling shape change. In this example, the change of shape also comprises stretching and / or compressing of the device 100a, 100b, 100c. This enables the device 100a, 100b, 100c to indicate distance or heading. However, distance or heading can be indicated in another manner in other examples, as will be described in more detail below. More generally, the shape change may be performed through bending and / or stretching in at least two degrees of freedom whilst the device 100a, 100b, 100c is held in the hand. For example, the device 100a, 100b, 100c may be able to bend left, right, up and / or down. This enables the device 100a, 100b, 100c to point in any direction. The device 100a, 100b, 100c may be able to stretch forwards and compress backwards. The device 100a, 100b, 100c can indicate a distance (for example to the object 104) or heading based on the extent of stretching and / or compression and / or in another manner. In this example, the device 100a, 100b, 100c comprises an elongate body, and the shape- change mechanism is operable to cause the elongate body of the device 100a, 100b, 100c to change shape by changing straightness and / or length. This enables intuitive pointing. In this example, the device 100a, 100b, 100c comprises a microcontroller (not visible) and the microcontroller is configured to control the shape-change mechanism based at least in part on the spatial sensor data. This can provide high-speed control when converting the spatial input into a shape-changing output. The control of the shape-change mechanism may also be based on IMU data. For example, IMU data may be used to perform tilt compensation such as described below with reference to Figure 8. The device 100a, 100b, 100c may notify the user 102a, 102b, 102c that the object 104 has been identified and / or that the user is pointing the device 100a, 100b, 100c towards the target position 104. Such a notification may be delivered when the user is pointing relatively near to the target position 104, for example within seven degrees. Such notification may be via a vibratory alert or otherwise. The alert may or may not be persistent. For example, a persistent alert could be a vibration or auditory signal which persists or persistently repeats whilst the device 100a, 100b, 100c remains pointed close to the target position 104. As such, the device 100a, 100b, 100c may be operable to provide a vibratory output. This enables the device 100a, 100b, 100c to provide alerts and other signals to the user 102a, 102b, 102c. In some examples, the device 100a, 100b, 100c and / or another device comprises a loudspeaker. In some examples, once the device 100a, 100b, 100c has directed the user 102a, 102b, 102c to point in a direction of interest, the device 100a, 100b, 100c verbally communicates the distance to the object of interest 104. For example, whilst pointed at a coffee cup, the device 100a, 100b, 100c could communicate “the cup is 60 centimetres away”. Such verbal communication may be an alternative to, or may be in addition to, the stretching and / or compression to indicate distance or heading and / or other haptic modalities. As such, in some examples, the device 100a, 100b, 100c is operable to communicate verbally a distance from the device 100a, 100b, 100c to the target position 104. Once the user 102a, 102b, 102c is pointed at the target position 104, in 3D space, two out of three dimensions are removed for locating the target position 104. The only remaining dimension is distance. If the device 100a, 100b, 100c is already pointed directly at the object 104, the device shape is straight. In this example, the shape-change mechanism is configured such that the device 100a, 100b, 100c is substantially straight when the device 100a, 100b, 100c is pointed at the target position 104 in the environment. This can be best seen in the device 100b in the second configuration. The term “substantially straight” is used herein to mean generally linear from the perspective of the user 102a, 102b, 102¢, for example as opposed to being curved or bent. The device 100a, 100b, 100c may therefore be substantially straight without being perfectly linear. This enables the device 100a, 100b, 100c to provide an unambiguous indication of 3D spatial direction or 3D object location. In this example, at least a portion of the device 100a, 100b, 100c is substantially cylindrical. Specifically, in this example, at least the gripping portion 110a, 110b, 110c of the device 100a, 100b, 100c is substantially cylindrical. The symmetry and the proprioceptive output enable the device 1003, 100b, 100c to be used in any orientation. The term “substantially cylindrical” is used herein to mean having generally parallel sides and having a generally circular cross-section. The device 100a, 100b, 100c may therefore be substantially cylindrical without being perfectly cylindrical. For example, as can be seen in Figures 1 and 2, the exposed shape-change mechanism portion 108a, 108b, 108c of the device 100a, 100b, 100c may result in a portion the device 100a, 100b, 100c deviating from being perfectly cylindrical while still being substantially cylindrical. Additionally, in this example, the back portion 106a, 106b, 106¢ and the front portion 112a, 112b, 112¢ of the device 100a, 100b, 100c are not cylindrical. In this example, the device 100a, 100b, 100c is operable to be controlled by at least one person other than the user 102a, 102b, 102c while the user 102a, 102b, 102c is using the device 100a, 100b, 100c. For example, the device 100a, 100b, 100c may be controlled by voice commands by the other person. The device 100a, 100b, 100c may alternatively or additionally be controlled by a device of the other person. The device of the other person may be an electronic device such as a computer, smartphone, tablet or another shape-changing haptic interface. For example, the other person may use a smartphone to identify the target position 104. This enables the other person, such as a remotely located person, to help the user 102a, 102b, 102c navigate. This will be explained in more detail below. Although, in this example, the device 100a, 100b, 100c analyses the camera frame, one or more other devices may perform such analysis, in addition to or as an alternative to the device 100a, 100b, 100c performing such analysis. For example, the device 100a, 100b, 100c may pair with a mobile computing device which performs the SLAM and computer vision computations, and / or provides other sensory input such as GPS co-ordinates or device orientation data. Examples of such mobile computing devices include, but are not limited to, smartphones, tablet computing devices, and laptop computing devices. The device 100a, 100b, 100c may pair with such other devices wirelessly, for example using Bluetooth™, or otherwise. The use of such other devices helps keeps the size and / or weight and / or costs of the device 100a, 100b, 100c low and enables over-the-air updates to the device 100a, 100b, 100c without the device 100a, 100b, 100c having capability to receive such updates directly. If the user 102a, 102b, 102c grips the device 100a, 100b, 100c tightly such that it cannot change shape, the user 102a, 102b, 102c may still feel the force indicating where the shape-change mechanism is trying to bend the device 100a, 100b, 100c towards, without damaging the shape- change mechanism. Referring to Figures 3 and 4, there is shown another example of a sensory substitution device 100. In this example, the gripping portion 110 has a single gripping region between the exposed shape-change mechanism portion 108 and the front portion 112. However, the user of the device 100 may nevertheless also use the rear portion 106 and / or the exposed shape-change mechanism portion 108 to grip the device 100. The rear portion 106 and / or the exposed shape-change mechanism portion 108 may therefore be considered to form part of the gripping portion 110. In this example, the shape changes are accomplished by three electrical motors 116 arranged on the front portion 112 of the device 100. In this example, the electrical motors 116 are equally spaced circumferentially around, and protrude outwardly from, the front portion 112 of the device 100. A different number of motors 116 can be used in other examples. The position of the motor(s) 116 can be different in other examples. In this example, each motor 116 variably tensions a respective one of three antagonistic tendons 118 which run up the length of the device 100. A different number of antagonistic tendons 118 can be used in other examples. In this example, each motor 116 drives a respective pulley 120. Only two pulleys 120 are shown in Figure 4, it being understood that three pulleys 120 are present in this specific example. In this example, each antagonistic tendon 118 is drawn in around its respective pulley 120 when its respective motor 116 is driven in one direction and is allowed to be drawn out from its respective pulley 120 as its respective motor 116 is driven in the other direction. Alternatively, all three pulleys 120 may draw in or pull out simultaneously to cause the device 100 to retract or extend along its length. More generally, at least two electrical motors 116 and at least two antagonistic tendons 118 may be used. However, if three independent antagonistic tendons 118 are used, the device 100 can compress and elongate, as well as bend. As such, in this example the shape-changing device 100 is driven by motors 116 and antagonistic tendons 118. This may be more compact, faster-moving and cheaper than other possible mechanisms. The motors 116 may readily be controlled by software. In this example, the device 100 comprises a shape-changing membrane, the surface of which corresponds to the external shape-change mechanism portion 108. In this example, the shape- changing membrane comprises alternating flexible vertebrae 122 and rigid vertebrae 124. In this example, the flexible and rigid vertebrae 122, 124 are ring-shaped. In this example, each antagonistic tendon 118 passes through a respective one of three holes 126 in each of the gripping portion 110, the flexible vertebrae 122, the rigid vertebrae 124, and the end portion 106. In Figure 4, one such hole is indicated with reference sign 126, it being understood that other holes are shown in Figure 4. This arrangement results in a device body which can change shape in three degrees of freedom. In particular, the device body can bend left, right, up and down, and extend forwards and backwards, or some combination of each degree. In this example, the flexible vertebrae 122, and hence the shape-change mechanism, comprises elastomeric material. The elastomeric material enables the shape-changing membrane to compress, expand, and bend. The elastomeric material also reduces the likelihood of, or even prevents, the shape-change mechanism from pinching the hand of the user when the device 100 changes shape. In examples, the device 100 comprises a sheath to cover the shape-change mechanism 108. The sheath may be corrugated. The sheath may prevent the user from being pinched while still allowing movement of the shape-change mechanism. The sheath may prevent ingress of outside material into the shape-change mechanism, reducing maintenance and increasing longevity. The sheath may play a mechanical role in the shape-change mechanism. The sheath may, alternatively, cover the shape-change mechanism. In this example, the motors 116, antagonistic tendons 118, flexible vertebrae 122 and rigid vertebrae 124 all form part of the shape-change mechanism. As can best be seen in Figure 4, in this example, the end portion 106, the gripping portion 110, the flexible vertebrae 122 and the rigid vertebrae 124 are hollow. This provides a lightweight device 100, which uses relatively small amount of material, while still being sufficiently rigid for the intended use. The hollow centre of the device 100 can comprise other components of the device 100. In this example, the shape-change mechanism is configured to cause the device 100 to return to an equilibrium shape when the device 100 is powered off. In this example, when the device 100 is powered off, the elastomeric material relaxes to its equilibrium state, and the motors 116 allow the antagonistic tendons 118 to unwind from the pulleys 120 such that the device 100 returns to being straight. The device 100 has, in effect, a self-righting design, which returns the device 100 to the straight, equilibrium configuration when the device 100 is powered off. This enables the device 100 to be more ergonomic and the device 100 better fits in a pocket of the user if the device 100 runs out of power. The device 100 may comprise one or more end-stop sensors to align the straight position. The end-stop may be a “home” position. The device 100 may comprise a locking mechanism to fix the device 100 in the home position when not in use. In some examples, three or more independently actuated tendons 118 are used. This enables the tendons 118 to be automatically tensioned. This also enables ease of assembly. In some examples, two or more tendons 118 with two or more independent actuators are used. This enables non-plane-constrained shape-change. In some examples, the pulleys 120 are deeply grooved pulleys 120. Such pulleys 120 may be used in combination with fixed-length tendons 118. This may reduce or remove the possibility of the tendons 118 slipping off the pulleys 120. In some examples, the device 100 comprises low-friction sleeving. An example of low-friction sleeving is a polytetrafluoroethylene (PTFE) tube. Tendons 118 may be routed through such low- friction sleeving. The sleeving diameter may be substantially the same as the diameter of the antagonistic tendon 118 to limit play. The device 100 may comprise a rotationally constraining linkage. The rotationally constraining linkage restricts the device 100 to having no torsional degrees of freedom. The rotationally constraining linkage may surround the actuator(s). Alternatively, the rotationally constraining linkage may be surrounded by the actuator(s). A stiff sheath may double as such a linkage by being bendable but not twistable. Such a sheath may be made of rubber, for example. In more detail, in examples, the shape-change mechanism is linked by a joint that does not allow rotation but enables pan and tilt motions. Examples of such a joint include, but are not limited to, a universal linkage and a flexible shaft. Referring to Figure 5, there is shown another example of a sensory substitution device 100. Figure 5 demonstrates how the device 100 can change shape with motion that is not plane- constrained, whereby to provide proprioceptive output to a user. In Figure 5, a first 2D plane is indicated with reference sign 126 and using broken lines, and a second 2D plane is indicated with reference sign 128 and using solid lines. In this example, the device 100 can change shape with motion that is not plane-constrained. In particular, the motion is not constrained to being in only one of the first or second planes 126, 128. Instead, the motion can be in either of the first or second planes 126, 128, or outside of the first and second planes 126, 128 in the 3D space surrounding the device 100. However, the device 100 may have plane-constrained motion in other examples. Referring to Figure 6, at item 200, an error is calculated as a difference between a desired pointing direction 202 and an actual pointing direction 204. In this example, the actual pointing direction 204 is determined by a device spatial sensor 206. In this example, a controller 208 controls a shape-change mechanism 210, for example a shape-changing mechanism, to change the shape of the device. At item 212, the user senses the device shape and, at item 214, moves their body based on the sensed device shape. The process represented in Figure 6 can be repeated until the actual pointing direction 204 matches the desired pointing position 202, at which point the output of error item 200 is zero. When this is achieved, the device is pointing at the target object(s) and / or location(s) and the shape of the device is substantially straight. In specific examples, the device shape updates twenty times per second, based on SLAM and / or computer vision input, and user movements. As such, in this example, the device is configured to change shape based on closed-loop feedback, where the closed-loop feedback is based on a difference between the target position 202 in the environment and a position 204 in the environment at which the device is pointed. The shape of the device may update multiple times per second, giving closed-loop feedback. Referring to Figure 7, there is shown another example of a sensory substitution device 100. In this example, the device 100 comprises a handheld body portion 130, a smartphone mount portion 132, and a smartphone portion 134. In this example, the handheld body portion 130 is generally rectangular cuboid in shape. In this example, the smartphone portion 134 comprises a smartphone 136. In this example, the smartphone 136 comprises the one or more spatial sensors 114. In this example, the handheld body portion 130 does not comprise any spatial sensors. In this example, the sensory substitution device 100 comprises the handheld body portion 130, the smartphone mount portion 132, and the smartphone portion 134. However, the sensory substitution device 100 may comprise fewer and / or different components in other examples. In accordance with this example, costs of the handheld body portion 130 can be kept low. Upgraded functionality may be enabled as new smartphones 136 are released. The handheld body portion 130 may have no sensors (such as GPS and IMU) on board and a camera 114 of the smartphone 136 may be used. It may also be easier to release and install software updates in this configuration. In another example, the shape-change mechanism of the sensory substitution device 100 is comprised in a handheld component 130 of the sensory substitution device 100 and the one or more spatial sensors 114 are comprised in one or more on-the-person components of the sensory substitution device 100. An on-the-person component may be a handheld component 130 (held by a person) or may be a wearable component (worn by a person). Examples of wearable components include, but are not limited to, headsets and smart glasses. Examples of such headsets include, but are not limited to, AR and virtual reality (VR) headsets. As such, the sensory substitution device 100 may be an on-the-person device having one or more handheld components 130 and one or more other on-the-person components, where an on- the-person component is, in use, on the person using the device 100. The on-the-person component may be on the person using the device 100 by being handheld or by being wearable. Each such component may have its own housing. Such a device configuration is operable in any environment in which the person is located without the need for external infrastructure, for example, Bluetooth™ beacons or availability of GPS reception, thus comprises a standalone sensory substitution device. In some examples described herein, a target position 104 is represented in the centre of a frame captured by a sensor 114 of the sensory substitution device 100. This may be the case where the sensor 114 is arranged in the pointing direction of the sensory substitution device 100. In other examples, the target position 104 may not be represented in the centre of a frame captured by a sensor 114 of the sensory substitution device 100. This may be the case where the sensor 114 is comprised in a wearable component and is not aligned with the pointing direction of the sensory substitution device 100. In such examples, the sensory substitution device 100 may be configured to calculate the pointing vector using, for example using a fiduciary marker on one or more of the on-the-person components or via other tracking by the spatial sensor(s) 114 even without fixed alignment between the handheld component 130 and the sensor 114. Referring to Figure 8, there is shown another example of a sensory substitution device in three different configurations. In Figure 8, the device is indicated with reference signs 100a, 100b, 100c in the first, second and third configurations respectively. This example demonstrates tilt compensation functionality, and assumes that a target position (not visible) is above the device 100a, 100b, 100c. Tilt compensation allows the user to move their wrist in a natural fashion without receiving incorrect cues. As such, the device 100a, 100b, 100c comprises a tilt compensator. The tilt compensator may be hardware-based and / or software-based. In the first configuration, the device 100a is at a 0-degree rotation, a target vector 138a points vertically upwards, and the device 100a is bent and pointed vertically upwards. In the second configuration, the device 100b is at a 45-degree rotation, a target vector 138b points vertically upwards again, and the device 100b is again bent and pointed vertically upwards with respect to the ground. Similarly, in the third configuration, the device 100c is at a 90-degree rotation, a target vector 138c points vertically upwards again, and the device 100c is again bent and pointed vertically upwards with respect to the ground. The rotation between the first, second and third configurations can readily be seen in the rotation of patterns 140a, 140b, and 140c on the front of the device 100a, 100b, 100¢c. As such, in this example, the device 100a, 100b, 100c is pointed vertically upwards with respect to the ground in each of the first, second and third configurations, and adjusts its shape change to remain pointed vertically upwards even when the device 100a, 100b, 100c is tilted or rotated. An IMU and / or a spatial sensor may be used to detect the tilt of the device 100a, 100b, 100c. Examples of spatial sensor include, but are not limited to a camera, a LIDAR sensor and an external virtual reality motion tracker. The device 100a, 100b, 100c is configured to adjust its shape-change motion accordingly. For example, the device 100a, 100b, 100c may determine an initial direction of shape change based on a 0-degree tilt angle and may transform that initial direction of shape change to derive a tilt-compensated direction of shape change based on the amount of tilt. Referring to Figure 9, there is shown another example of a sensory substitution device in five different configurations. In Figure 9, the device is indicated with reference signs 100a, 100b, 100c, 100d, 100e in the first, second, third, fourth and fifth configurations respectively. Figure 9 illustrates various bending directions of the example device 100a, 100b, 100c, 100d, 100e. In the first configuration, the device 100a is straight. In the second configuration, the device 100b is upward-bending from the perspective of the user. In the third configuration, the device 100c is rightward-bending from the perspective of the user. In the fourth configuration, the device 100d is downward-bending from the perspective of the user. In the fifth configuration, the device 100e is |leftward-bending from the perspective of the user. In this example, the device 100a, 100b, 100c, 100d, 100e comprises a first recess 142a. The recess may also be referred to as a “notch”. A recess may leverage other sensory modalities of a user, such as sliding sensations and the edge detection capacities of the human fingertips. In this example, the first recess 142a is in an upper surface of the device 100a, 100b, 100c, 100d, 100e. In this example, the first recess 142a is a thumb recess for a thumb of a user of the device 100a, 100b, 100c, 100d, 100e. In this example, the device 100a, 100b, 100c, 100d, 100e comprises a second recess 142b. In this example, the second recess 142b is in a lower surface of the device 100a, 100b, 100¢, 100d, 100e. In this example, the second recess 142b is a finger recess for a finger of a user of the device 100a, 100b, 100¢, 100d, 100e. In this example, the first and second recesses 142a, 142b are ergonomic recesses for the thumb and forefinger of the user. More generally, however, the first and second recesses 142a, 142b are for respective digits (of which a thumb and a finger are examples) of a hand of the user. As such, shape-change can, in effect, be used between the thumb and forefinger of a user while the user has a ‘handle’ which can be held in their palm. By having the shape-change at the front of the device 100a, 100b, 100c, 100d, 100e, the device 100a, 100b, 100c, 100d, 100e can be relatively small and mechanically simple. The device 100a, 100b, 100c, 100d, 100e may also comprise a rigid section to contain components. In this example, the device comprises a sensor suite 114. In this example, the sensor suite comprises a plurality of sensors 114. The plurality of sensors 114 may be comprise two or more sensors 114 of the same type as each other. The plurality of sensors 114 may be comprise two or more sensors 114 of different types from each other. In this example, the device 100a, 100b, 100c, 100d, 100e comprises a shape-change portion 108 comprising a shape-change mechanism. Although this example uses recesses, another type of tactile marker 142a, 142b may be used. For example, one or more ridges and / or triggers may be used in addition to, or as an alternative to, one or more recesses. Such tactile markers 142a, 142b encourage the user to hold and use the device 100a, 100b, 100¢, 100d, 100e in the correct way. One or more ridged sections may be combined with one or more shape-change sections. For example, multiple shape-change sections in combination with multiple rigid sections may allow the entire device 100a, 100b, 100c¢, 100d, 100e to bend along its length. Referring to Figure 10, there is shown another example of a sensory substitution device 100. In this example, the device 100 provides an additional momentary stimulus (or stimuli) other than shape-change when the device 100 is pointed within an error bound 143 of the (exact) target position 104. The target position 104 may be associated with a target vector. The momentary stimulus is represented schematically in Figure 10 by item 144. The stimulus 144 lets the user know they are pointed close to the target position 104. The stimulus 144 may be a continuous stimulus, such as a persistent vibration or audio alert. The stimulus 144 may be a temporal pattern of alerts, such as beeping or a vibration pattern. For example, the device 100 may be configured such that, when the device 100 is pointed within a certain number of degrees, for example seven, of the target position 104, the momentary stimulus 144 is activated. This informs the user that they are pointing close to the target position 104. In another example, the device 100 provides a momentary stimulus 144 (or stimuli) when the device 100 is pointed outside a given bound 143 of the (exact) target position 104. This may provide a warning that they are pointing outside the bound 143. The device 100 may provide one type of stimulus 144 if the device 100 is pointed within a given error bound 143 and another type of stimulus 144 if the device 100 is pointed outside the or another given error bound 143. In this example, the stimulus 144 comprises vibration. As such, the device 100 may start throbbing based on the closeness of the pointing of the user to the target position 104. However, vibration may numb the hand of the user and may be distracting. This vibration may take any frequency, magnitude or waveform and may be dynamic or transient in any of these aspects. The stimulus 144 may comprise illumination. Some vision impaired users may be able to sense bright light, either of a certain colour or of any colour. More generally, the device 100 may incorporate programable lighting to make use of the remaining vision of the user. It is surprising that this is effective for blind users, namely users with severe visual impairment. As an alternative to, or in addition to, non-audible feedback, the stimulus 144 may comprise an audio cue. The use of the momentary stimulus 144 drastically improves device performance by enabling a user to recognise easily when they have or have not acquired a target vector. The user can then, in effect, stop focusing on trying to locate the object 104, reducing cognitive load. Otherwise, there may be ambiguities as to whether or not the object or target direction 104 has been located. Referring to Figure 11, there is shown an example sensory substitution device 100 providing positive feedback to a user 102. Such feedback may be considered to be ‘positive’ in that the device 100 positively points the user 102 along a path 146 the user 102 is to follow. The path may be a virtual path in that the path is not depicted in the real world. In this example, the user 102 navigates along the path 146 to a destination 148 using the device 100. In this example, the path 146 is an optimal path and enables the user 102 to avoid multiple obstacles 150a, 150b in the environment. As such, the user 102 can use the device 100 to follow the path 146 whilst simultaneously avoiding obstacles 150a, 150b. The device 100 may use one or more onboard sensors to detect the obstacles 150a, 150b and / or undertake-path planning. Examples of such sensors include, but are not limited to, GPS, LIDAR, digital imaging and RADAR. The device 100 may use an additional haptic modality to indicate to the user 102 to move along a target vector. As such, the device 100 may indicate to the user 102 where to go, rather than where not to go. This differs from ‘negative’ feedback where, for example, the device 100 may vibrate if danger or potential danger is detected. The device 100 may therefore be considered to be a device of guidance rather than a device of pure exclusion. A cane is, by contrast, a device of exclusion. The example device 100 may, for example, positively guide the user 102 across a field, which is not the purpose of a cane. Referring to Figure 12, there is shown an example sensory substitution device providing negative feedback to a user. Such feedback may be considered to be ‘negative’ in that the device warns the user of a danger or potential danger. In Figure 12, the user is shown in three different positions and the device is shown in three different respective configurations. In Figure 12, the device is indicated with reference signs 100a, 100b, 100c in the first, second and third configurations respectively, and the user is indicated with reference signs 102a, 102b, 102c in the first, second and third positions respectively. In this example, the environment comprises obstacles in the form of corridor walls or “edges” 150a, 150b. This example shows how the user 102a, 102b, 102c can use the device 100a, 100b, 100c as an intuitive tool to detect the edges 150a, 150b of the corridor whilst navigating. When pointed at a wall or other obstacle, 150a, 150b, the device 100a, 100b, 100c deflects from the obstacle 150a, 150b to let the user 1023, 102b, 102c know of the presence of the obstacle 150a, 150b. This is a different mode from positive feedback. This can also provide, in effect, a virtual cane. A virtual cane mimics a physical cane. However, there can be additional features of a virtual cane. For example, the ‘virtual length’ of the virtual cane may be adjusted. In other words, the distance from an obstacle or other object at which negative feedback is provided to the user may be adjusted. Initially, the distance may be relatively long and, over time, the distance may be shortened. Different distances may be used based on the risk associated with the obstacle in question, the competency of the user 102 in using the device 100a, 100b, 100c, or otherwise. A virtual cane may be more discrete than a physical cane. A virtual cane may be easier to store and transport than a physical cane. A virtual cane could still be used with a physical cane, for example attached to a physical cane if desired. The device 100a, 100b, 100c may be selectively operable in the negative feedback mode, for example based on user input. In more detail, in this example, in the first configuration, the device 100a is bent to the right, and the front portion is parallel to both walls 150a, 150b. In the second configuration, the device 100b is straight and, again, the front portion is parallel to both walls 150a, 150b. In the third configuration, the device 100c is bent to the left and, again, the front portion is parallel to both walls 150a, 150b. Referring to Figure 13, there is shown another example of a sensory substitution device 100. In this example, the device 100 comprises a hand strap 152. The hand strap 152 is connected, at both ends, to the device 100. In use, the user 102 inserts their hand between the hand strap 152 and the main body of the device 100 such that the device 100 is in the palm of the hand and the strap 152 is on the back of the hand. This enables the user 102 to move their fingers freely and feel the shape of the device 100 while still supporting the device 100. As depicted in Figure 13, when using the hand strap 152, the user 102 may still firmly grip the device 100, may touch the device 100 using their thumb and forefinger, or may release their grip of the device 100 entirely. Using just a thumb and forefinger can provide high resolution haptic communication with the user in view of the high sensitivity of the thumb and forefinger. Releasing grip altogether whilst having the hand strap 152 engaged enables hands-free operation, enabling, for example, for the user pick up or manipulate environmental objects with the same hand the device 100 is strapped to. The hand strap 152 may be tight around the main part of the hand of the user 102 once the user 102 has slipped their hand through the strap 152. As explained above, the user 102 does not need to grip the device 100 in such situations. As such, the strap 152 can eliminate the need for the user 102 to support the device 100 while still enabling the user 102 to feel the shape of the device 100. As such, the user 102 does not need to hold the device 100, but can feel the device 100 knowing that the device 100 is supported, such that the device 100 almost ‘floats’ on the hand of the user 102. This also enables the user 102 to have their fingers free. This may enable them to push buttons on an elevator, to carry a cane, and so on. These benefits are surprising on a small, handheld device. The strap 152 may have an adjustable tightness. Referring to Figure 14, there is shown another example of a sensory substitution device 100. In this example, the device 100 is attached to a cane 154 of the user 102. The device 100 may be clipped onto the cane 154, for example. Referring to Figure 15, there is shown an example of a system 300 comprising an example sensory substitution device 100. This example relates to teleassistance, which may also be referred to as “remote assistance”. In this example, the system 300 comprises a sensory substitution device 100 and a remote device 156. In this example, the sensory substitution device 100 is used by a user 102 and the remote device 156 is used by a remote entity 158. The user 102 may be referred to as the “assisted”, “assisted person”, or the like. The remote entity 158 may be a human or otherwise. The remote entity 158 may be referred to as an “assister”, “assisting person”, or the like. The assisting person 158 may be a person on the remote end of a video feed who is providing teleassistance to the assisted person 102. The assisted person 102 may be a person sending a video feed to the device 156 of the assisting person 158, and who will follow guidance provided by the assisting person 158. The degree of remoteness, which is depicted schematically by line 160 in Figure 15, can differ in different examples. The remote entity 158 may be in the same room as the user 102, a different room in the same building as the user 102, or much further away from the user 102. As such, “remote” should be understood broadly to mean not in exactly the same location as the user 102. In this example, teleassistance is provided via digital markers. In more detail, in examples, teleassistance is provided by having the assisting person 158 located in one place select a location to place a digital marker within the local environment of the assisted person 102 based on a live video feed from the location of the assisted person 102. The person 102 being assisted is then guided towards the digital marker using a non-visual interface (for example as described herein), audio and / or visual cues. This leverages human input to help people locate objects and / or navigate spaces, which may be superior to completely automated solutions. Video calls can currently be used to aid a person who is in a different place from an assisting person. Typically, the assisting person and the person being assisted communicate using a mobile phone or smart glasses, with verbal instructions being provided in response to input from a live video feed. For example, a sighted helper may aid a vision impaired person to identify and / or locate objects, and navigate environments. However, using verbal communication alone through a video call is highly ambiguous. Additionally, if the person being assisted moves the line of sight of the camera out of the location to which the assister is trying to direct the assisted, then the assisted will first need to be directed to point the camera in a particular direction before any useful guidance can be provided. In examples, the assister 158 uses a remote video feed sent from the device 100 of the assisted user 102 to place a locational anchor in 3D space within the local environment of the assisted user 102. For example, the assister 158 may tap their finger on the appropriate screen location in a video feed displayed on the remote device 156. The assister 158 may be able to pause the video feed from the assisted user 102. This gives the assister 158 time to comprehend the footage and place a digital marker accurately. The device 100 of the assisted person 102 and / or another device then calculates where the digital marker was placed, even if the assisted person 102 has moved from the original space. The locational digital marker may thereby be placed in a position in 3D space within the local environment of the assisted user 102 by another person 158 who may be located anywhere in the world. The locational digital marker may be used to indicate a direction for navigation. The digital marker may automatically update its position in the environment as the assisted user 102 moves. As such, the assisted user 102 can be guided along a particular directional vector without further assistance from the assister 158. If a digital marker is placed on an object, object detection inference can be used to identify automatically what the object is. Conversely, object detection can be used to identify an object and place a digital marker on the identified object. When the digital marker has been placed, the assisted user 102 may be guided towards the digital marker using a haptic interface, automated audio cues and / or visual cues from their device 100. In more detail, the person seeking assistance, namely the assisted person 102, may use the device 100, to initiate or otherwise establish a video call with another person, namely the assister 158, who may be located anywhere in the world. The device 100 may comprise a smartphone, smart glasses, and / or a sensory substitution device such as described herein. The assisting person 158 sees the live video stream from the device 100 of the assisted user 102 and can tap their finger on the screen of their own device 156 to place an digital marker within the local environment of the assisted user 102. The device 156 of the assisting person 158 may be a smartphone, tablet computing device, laptop computer, desktop computer, etc. The digital marker is registered by the device 100 of the assisted user 102. The device 100 may track its own location within 3D space. The device 100 may be used to draw the attention of the user 102 towards the digital marker, for example using audio, visual and / or haptic cues. Such examples may be used to help the assisted user 102 navigate spaces, identify objects and / or identify other points of interest. In a specific implementation, the assisting person 158 can pause and / or rewind the video feed. In the meantime, the device 100 of the assisted person 102 is tracking its movements while the video is paused and / or otherwise being reviewed. The assister 158 can then review the footage and place a digital marker in a place in which the assisted user 102 was recently located As such a vector may be communicated by directing a user to point along a target vector. The pointing gesture is a universal hand signal to indicate a location that is observed even in congenitally blind children. This provides improved efficacy compared to just touch alone. Referring to Figure 16, there is shown an example in which the device 100 is pointing towards a target 104 such that an extrapolation of the device centre line 162 is approximately aligned with the target 104. The device centre line 162 need not be straight. For example, the device 100 may have a curved or other ergonomic shape. In this example, the device 100 works to align the centre line 162 with the target 104. The device 100 achieves this by changing shape such that, when returned to the home position, the centre line 162 will align with the target 104. In this example, the centre line 162 of the device 100 is along the longest axis of the device 100. In this example the sensory substitution device 100 includes a handheld shape-change mechanism. Various additional examples will now be described. In some examples, the user 102a, 102b, 102c can direct the device 100, 100a, 100b, 100c, 100d, 100e to navigate to locations 104 using only their voice though natural language processing. The device 100, 100a, 100b, 100c, 100d, 100e pairs with another device, such as a smartphone. Applications, such as mapping applications, are also interfaced. The user 102a, 102b, 102¢ walks with the device 100, 100a, 100b, 100c, 100d, 100e, holding the device 100, 100a, 100b, 100c, 100d, 100e forward-facing like a flashlight. In examples, the forward-facing camera 114, 114a, 114b, 114c feeds video to a computer vision application to identify important navigational features 104 which are typically identified by a guide dog. Examples of such navigational features include, but are not limited to, pavements and zebra crossings. The user 102a, 102b, 102c is kept from walking into a road by navigational guidance though the touch sense. In some examples, the user 102a, 102b, 102c can use the device 100, 100a, 100b, 100c, 100d, 100e to ‘read’ text using text-to-speech functionality. For example, the user 102a, 102b, 102c could aim the device 100, 100a, 100b, 100c, 100d, 100e at a paragraph of text and use their voice to command the device 100, 100a, 100b, 100¢, 100d, 100e to read the text. In some examples, the user 102a, 102b, 102c can use the forward-facing camera 114, 114a, 114b, 114c on the device 100, 100a, 100b, 100c, 100d, 100e to magnify items. Images of such items can be streamed to a display, such as a smartphone screen or smart glasses display, for closer inspection by the user 102a, 102b, 102c. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e can be used in a mode in which the user 102a, 102b, 102c aims the device 100, 100a, 100b, 100c, 100d, 100e at an object of interest 104, initiates a command (for example by pressing a button and / or issuing a voice command), and the device 100, 100a, 100b, 100c, 100d, 100e calls out the largest object 104 identified in the camera frame. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e does not have any sensors 114, 114a, 114b, 114c, such as forward-facing sensors 114, 114a, 114b, 114c. Instead, the device 100, 100a, 100b, 100c, 100d, 100e is a shape-changing, human-machine interface that can be used to convey directional information based on sensory information provided by other technologies. Examples of such other technologies include, but are not limited to, a smartphone, Bluetooth™ beacons, smart glasses, GPS, or wearable technology. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e is handheld in that the weight of the device 100, 100a, 100b, 100c, 100d, 100e is held off the ground only by the hand of the user 102a, 102b, 102¢c. However, the device 100, 100a, 100b, 100c, 100d, 100e may be supported by another object in other examples. Examples of other such objects include, but are not limited to, guide canes, walking sticks and dog leashes. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e is used by a vision impaired user 102a, 102b, 102c. However, the device 100, 100a, 100b, 100c, 100d, 100e may be used by a person who is not vision impaired. For example, the device 100, 100a, 100b, 100c, 100d, 100e could be used in a warehouse to help a user who may or may not be vision impaired locate items more efficiently than they would be able to do without such a device 100, 100a, 100b, 100c, 100d, 100e. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e is configured to perform non- linear actuation. In such examples, an amount of shape change of the device 100, 100a, 100b, 100c, 100d, 100e correlates non-linearly to an amount of error between a pointing position and a target position 104. For example, the rate (or other amount) of change of actuation with respect to error, at small errors, may be greater than the rate (or other amount) of change of actuation with respect to error, at higher errors. The error may correspond to the difference between the desired pointing direction and the actual pointing direction, such as described above with reference to Figure 6. This may improve the pointing accuracy of the user 102, 102a, 102b, 102c. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e is optimised for the user 102, 102a, 102b, 102c. This may use machine learning or otherwise. As such different devices 100, 100a, 100b, 100c, 100d, 100e may be tuned for each user to have per-user critical damping. Such tuning may be based on temporal usage. Such tuning may use a Proportional-Integral-Derivative (PID) controller to reduce overshoot and, hence, increase the likelihood of the user 102, 102a, 102b, 102¢c promptly pointing at the desired target position 104. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e is optimised for an intuitive path of motion. For example, the device 100, 100a, 100b, 100c, 100d, 100e may be configured to prevent a user 102, 102a, 102b, 102c from being directed to move the device 100, 100a, 100b, 100c, 100d, 100e over their head. Even if a path to a target position 104 that involves the device 100, 100a, 100b, 100c, 100d, 100e being moved over the head of a user 102, 102a, 102b, 102c is a mathematically and / or spatially optimal path, it may not be a biomechanically optimal path. In more detail, the most efficient path (mathematically and / or spatially) may feel unnatural for a user 102, 102a, 102b, 102c. As explained above, if a target vector is located behind a user 102, 102a, 102b, 102c, then the device 100, 100a, 100b, 100c, 100d, 100e may otherwise bend upwards to direct the user 102, 102a, 102b, 102c to move the device 100, 100a, 100b, 100c, 100d, 100e in a path that is over their head, rather than bending to the left or right to indicate to the user 102, 102a, 102b, 102c to turn around. In contrast, by directing the user 102, 102a, 102b, 102c on the biomechanically optimal path, more natural guidance may be provided. In some examples, one or more portions of the device 100, 100a, 100b, 100c, 100d, 100e other than the front of the device 100, 100a, 100b, 100c, 100d, 100e comprise one or more sensors 114, 1143, 114b, 114c. As such, sensors 114, 114a, 114b, 114c may be placed at strategic geometric locations around the device 100, 100a, 100b, 100c, 100d, 100e. One or more sensors 114, 114a, 114b, 114c may still be at the front of the device 100, 100a, 100b, 100c, 100d, 100e in such examples. This can optimise performance of SLAM, odometry and detection. Such sensors 114, 114a, 114b, 114c may take various forms. For example, such sensors 114, 114a, 114b, 114c may comprise cameras, including but not limited to low-resolution cameras, infrared cameras, and so on. This facilitates ‘inside-out’ odometry, namely determination of movement relative to a previous position. For example, a sensor 114, 114a, 114b, 114c may point downwards and backwards relative to the device 100, 100a, 100b, 100c, 100d, 100e, so as to sense an area directly in front of the user 102, 102a, 102b, 102c. Such sensor(s) 114, 114a, 114b, 114c may be placed such that the hand of the user does not occlude or limit the field of view of the sensor(s) 114, 114a, 114b, 114c. Such placement may be such that the field of view of the sensor(s) 114, 114a, 114b, 114c is not occluded by either a right hand or a left hand of a user 102, 102a, 102b, 102c, for ambidextrous usage. More generally, the device 100, 100a, 100b, 100c¢, 100d, 100e may have an ambidextrous design configured for ambidextrous usage. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e has a curved handle. This may fit the hand of the user 102, 102a, 102b, 102c more ergonomically than with a straight device. A straight handle may push on the thumb of the user 102, 102a, 102b, 102c, but can lead to the device having the same feel as a cane. In some examples, the user 102, 102a, 102b, 102c is trained to point along arbitrary target vectors through an automated induction game. This may involve prompting the user 102, 102a, 102b, 102c to point to a real-world target and / or to a synthetic (or ‘virtual’) target. This may also be used to calibrate device parameters or coefficients for the particular user 102, 102a, 102b, 102c¢ in question. Calibration may, alternatively or additionally, involve identifying which colours, if any, the user 102, 102a, 102b, 102c can see particularly well. Light of such colour(s) may be used to provide feedback to blind and partially sighted users 102, 102a, 102b, 102c as described herein. In some examples, the weight of the device 100, 100a, 100b, 100c, 100d, 100e is balanced such that the user 102, 102a, 102b, 102c can hold the device 100, 100a, 100b, 100c, 100d, 100e in their hand without needing to support the shape-changing portion of the device 100, 100a, 100b, 100c, 100d, 100e. In particular, the centre of mass of the device 100, 100a, 100b, 100c, 100d, 100e may balance in the palm of the hand of the user 102, 102a, 102b, 102c. If the user 102, 102a, 102b, 102c has to exert an effort to support the device 100, 100a, 100b, 100c, 100d, 100e, in particular the shape-changing portion, it may be more difficult for the user 102, 102a, 102b, 102c to sense the device shape and changes to the device shape. In some examples, the user 102, 102a, 102b, 102c can switch between objects identified in a camera frame by use of user input. Such user input may be active user input such as, but not limited to, speech. For example, the device 100, 100a, 100b, 100c, 100d, 100e may lock (or ‘latch’) onto an identified object. The device 100, 100a, 100b, 100c, 100d, 100e may then ‘flick’ between the identified object and one more other identified objects. For example, if the user 102, 102a, 102b, 102c is looking for a coffee mug, there may be multiple instances detected within one camera frame. The device 100, 100a, 100b, 100¢, 100d, 100e may prompt the user 102, 102a, 102b, 102c to point to the nearest coffee mug to the user 102, 102a, 102b, 102c. However, the user 102, 102a, 102b, 102c may be able to move the device 100, 100a, 100b, 100c, 100d, 100e to lock onto a different coffee mug. Hysteresis may be used to limit the extent to which the device 100, 100a, 100b, 100c, 100d, 100e ‘flicks’ between different identified objects. In some examples, cloud-based object recognition may be used. This may reduce computational burden on the device 100, 100a, 100b, 100¢, 100d, 100e. One or more digital markers may be set at the same time as sending camera frame data to the cloud. For example, for a given frame of video, digital markers may be set in the middle of bounding boxes around all identified objects in the frame. Cloud-based object detection inference may be used and the device 100, 100a, 100b, 100¢, 100d, 100e may account for the processing time delay. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e is configured to use coded pulses of vibration to communicate, for example to deaf-blind persons. For example, the device 100, 100a, 100b, 100c, 100d, 100e may be used in this manner to communicate text. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e comprises one or more absolute encoders. This can reduce or eliminate the need for a start-up procedure to know the device shape-change mechanism position. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e comprises one or more piston-based actuators. Pistons push, whereas tendons pull. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e is operable to generate an alert to enable the user 102, 102a, 102b, 102c to locate the device 100, 100a, 100b, 100c, 100d, 100e. The alert may be an audio alarm, a vibration alert, or otherwise. In some examples, a target position 104 may correspond to a marker. An example of a marker is a pre-existing, persistent digital marker. Such markers may have been ‘placed’ on a map of an environment prior to the user 102, 102a, 102b, 102c being in the environment. The marker positions may correspond to points of interest in the environment. Such markers may correspond to digital markers. Such markers may correspond to fiducial markers. Such markers may be placed manually, prior to the user 102, 102a, 102b, 102c being in the environment concerned. Such markers may therefore differ from automatically recognised markers and markers placed by teleassistance. In some examples, digital marker points are used to interpolate a path through an environment. This enables 3D spatial guidance. As explained above, an AR anchor is a type of digital marker. However, other types of digital marker may be used. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e is configured to project such a digital marker into the real world in a position corresponding to that of the digital marker. Such projection may involve a ‘goes before optics’ (gobo) projector. A gobo projector may project a design, such as a logo, onto a surface. In some examples, the device 100, 100a, 100b, 100c, 100d, 100e comprises one or more removable and / or interchangeable components. For example, the device 100, 100a, 100b, 100c, 100d, 100e may comprise one or more removable and / or interchangeable input components. Examples of such components include, but are not limited to, joypads and triggers. Another example of a removable and / or interchangeable component is a removable and / or interchangeable head. For example, different heads may comprise different sensor suites. As another example, one or more removable and / or interchangeable components may provide the device 100, 100a, 100b, 100c, 100d, 100e with waterproof functionality whereas the device 100, 100a, 100b, 100c, 100d, 100e may not be waterproof without such component(s). In some examples, the device 100, 100a, 100b, 100c, 100d, 100e is water-resistant. For example, the shape-change mechanism may be sealed such that the shape-change mechanism is itself water-resistant. In some examples described above, shape-change is used to indicate distance. In other examples, a secondary modality (other than shape-change) is used to indicate distance and may, surprisingly, be more effective at indicating distance than shape-change. For example, a small bump moving over the thumb or forefinger of the user 102, 102a, 102b, 102c may provide a haptic modality. In some examples, distance-based feedback is continuous. For example, temporal vibration may be used to indicate distance, with the vibrations getting faster and slower as the user 102, 102a, 102b, 102c gets closer to and further from the target position 104 respectively. In other examples, distance- based feedback may be persistent or intermittent. For example, the distance-based feedback may be activated when the user 102, 102a, 102b, 102c gets within a threshold distance of the target position 104 and may be deactivated when the user 102, 102a, 102b, 102c is outside the or another threshold distance from the target position. The secondary modality may use an oscillator to provide a pulling or tugging sensation. Such sensation may be based on a pulling vector. The oscillator may be in the front of the device 100, 100a, 100b, 100c, 100d, 100e. In another example, an inverted trigger may indicate distance. For example, the device 100, 100a, 100b, 100c, 100d, 100e may cause an inverted trigger to extend from and retract towards the device 100, 100a, 100b, 100c, 100d, 100e as distance to the target position 104 increases and decreases respectively, or vice versa. In some examples described above, a shape-change mechanism provides kinaesthetic output to a user 102, 102a, 102b, 102c¢ of the device 100, 100a, 100b, 100c, 100d, 100e. Another type of force-based mechanism may be used in other examples. Examples of other types of force-based mechanism include, but are not limited to, gyroscopes and asymmetrically excited linear oscillators. As such, a shape-change mechanism and / or another type of force-based mechanism (of which gyroscopes and asymmetrically excited linear oscillators are specific examples) may be used. In examples, such a force-based mechanism does not use vibration or audio to provide the kinaesthetic output and, as such, may be considered to provide non-vibratory, non-audio feedback. In some examples, the user 102, 102a, 102b, 102c may use a holster with a pouch to carry the device 100, 100a, 100b, 100c, 100d, 100e when not in use. Alternatively, the device 100, 100a, 100b, 100c, 100d, 100e may be carried in a different manner when not in use. For example, the user 102, 102a, 102b, 102c may wear a belt to which the device 102, 102a, 102b, 102c can be attached magnetically. Defining And Following Paths Examples will now be described in relation to defining and following paths. One or more features in connection with such examples may be used with any one or more other features described herein, unless such use would render one or more of such features technically inoperable. As explained above, existing applications, such as smartphone applications, for helping vision impaired people navigate are designed to convey information by audio or vibration output. Neither of these modalities can communicate 3D spatial information in a fast and accurate manner. As such, existing solutions only provide high-level navigation using sensors such as GPS. One existing system uses high-resolution LIDAR scans in conjunction with a smartphone camera to localise a user accurately. However, this requires scanning of locations before such locations can be navigated by the user. In particular, none of the existing solutions combines high-level navigation with low-level obstacle avoidance. In contrast, various examples described herein provide high accuracy path-tracing with obstacle avoidance. Some existing technologies for autonomous navigation through transport stations rely on strategic placement of Bluetooth™ beacons and / or fiducial markers. Again, these technologies have insufficient resolution, are ineffective at accounting for local obstacles, in particular temporary obstacles, and require significant changes to infrastructure. Most existing mapping software creates a map using a device and / or information source other than the navigation guidance device that will provide navigation guidance using the map. For example, such mapping software may use satellite data or high-resolution LIDAR scans. Very few users can make maps in this manner, for example because of the level of expertise and investment required. With existing systems, a path can be defined based on a map. However, navigation at a centimetre-accurate level using map-based paths requires very large and very detailed maps. Even when following a path on existing low-resolution maps, only a fraction of the mapping data available is used. The full mapping data, for example a world map, is too large to be stored on many smartphones for example and is, instead, streamed via the Internet. Such streaming of ultra-high- resolution maps, as would be used for centimetre-accurate map-based paths, is likely untenable in practice. Further, navigation may not be available in some environments, for example when underground and when the map has not been pre-loaded. Further, existing systems do not account for certain obstacles, such as in train stations, or at least make such obstacles difficult to avoid. Existing systems also cannot identify, with any certainty, whether or not a particular path is appropriate for a person, particularly a vision impaired person, to walk along. For example, some existing mapping systems suggest walking paths along busy main roads with no pavements. Similarly, some existing GPS-based mapping systems automatically generate paths that go through lakes or rivers. Navigation devices for vision impaired people should be highly dependable since they may be the only guide for vision impaired people. As such, defining paths based on maps generated by an automated route planner, such as Google Maps™, can lead to dangerous paths. However, even if a path is walkable, transient obstacles, such as pedestrians, may justify subtle, or even significant, deviations from the path. Various terms will be used in conjunction with examples below relating to defining and following paths. Such terms will now be explained. The term “device” is used herein to mean any hardware and / or software. A device may comprise one or more components. Where a device comprises multiple components, components may be in one or more housings. The term “navigation guidance device” is used herein to mean any device that is used in providing navigation guidance. A navigation guidance device may actively guide a user. A navigation guidance device may generate, obtain, process or otherwise use data that may be used to guide a user. For example, a device that generates and / or processes data that may be used to guide a user may be considered to be a navigation guidance device even if that device does not itself actively guide a user. A navigation guidance device may function in conjunction with one or more other navigation guidance devices to guide a user. For example, a first navigation guidance device may define a path to be followed by a user, a second navigation guidance device may receive, from the first navigation guidance device, data representing the path, and a third navigation guidance device may receive, from the second navigation guidance device, the data representing the path and actively guide the user using the received data. Navigation guidance devices may take various forms. For example, a navigation guidance device may comprise a smartphone, a tablet computing device, a server, a smart cane, a wearable device, a sensory substitution device as described above, or otherwise. The term “componentry” is used herein to mean one or more hardware and / or software components. The term “mechanism” is used herein to mean any hardware and / or software to perform a given function. The term “kinaesthetic output mechanism” is used herein to mean any mechanism that can provide output in a manner that can influence kinaesthesia. Examples of kinaesthetic output mechanisms include, but are not limited, to shape-change mechanisms and gyroscopes. A kinaesthetic output mechanism may convey tactile and proprioceptive feedback during a user’s interaction with a device comprising a kinaesthetic output mechanism. The term “obtain” is used herein to encompass any way of getting. Obtaining data, for example, may involve receiving data, generating data, retrieving data, or otherwise. The term “route” is used herein to mean a planned journey to a destination. The term “route data” is used herein to mean any data that is associated with a route. The term “path” is used herein to mean a specific way of reaching the destination of a planned journey which is traversable by a pedestrian. In particular, the term “path” is not being used to mean a pathway. For example, a “path” as used herein, may not involve any existing pathways. A route may be associated with one or multiple paths. For example, one path may provide a specific way of reaching a destination and another path may provide another specific way of reaching the same destination. A path may be associated with one or multiple routes. For example, a path may be re- used for multiple different routes. The term “path data” is used herein to mean any data that is associated with a path. “Path data” may be referred to herein as “path marker data”, “digital path marker data” or the like. The term “path-defining person” is used herein to mean any person that defines at least part of a path. A path-defining person may be referred to herein as a “user”. A path-defining person may define at least part of a path actively and / or passively. Actively defining a path may comprise the path- defining person defining one or more digital path markers associated with the path themselves, for example by interacting with a user device to define a digital path marker. Passively defining a path may comprise monitoring activity of the path-defining person and defining a digital path marker based on the activity. As such, a path-defining person may or may not knowingly or consciously define at least part of a path. The term “path-defining entity” is used herein to mean any entity that defines at least part of a path. An example of a “path-defining entity” is a path-defining person. Another example is a guide dog. The term “defining” is used herein, in relation to path-defining, and the like, to mean specifying. Terms such as “placing”, “recording”, “logging”, “creating” and the like may be used similarly herein. The term “path-following person” is used herein to mean any person that follows at least part of a path. A path-following person may be referred to herein as a “user”. The term “following” is used herein, in relation to path-following, and the like, to mean being guided along. Terms such as “tracing”, “retracing”, and the like may be used similarly herein. The term “defined path” is used herein to mean a path that has already been defined by a path-defining person. The term “walked path” is used herein to mean a path along which a path-defining person has walked. Walking along a path includes assisted walking. Assisted walking may comprise use of a mobility aid, a human guide, a guide dog, or otherwise. Walking includes all forms of self-propelled pedestrian locomotion, such as running and jogging, and inclusive options such as wheelchairs. In accordance with examples, walked paths provide increased safety and / or traversability compared to paths that have not been walked. For example, a plausible walking path could be defined between an origin and a destination based on existing mapping systems. However, that path may include an obstacle that is not recorded in the existing mapping system. Such a path may therefore not be safe, in particular, but not exclusively, to a vision impaired person. By providing guided navigation based on a walked path, examples described herein can provided enhanced navigation safety for pedestrians, such as vision impaired people. The term “pedestrian” is used herein to mean a person that navigates along a path. Such a person may be assisted when navigating. A pedestrian, as used herein, encompasses a person who uses a mobility aid such as a wheelchair, a mobility scooter or otherwise. The term “digital route marker” is used herein to mean a digital representation useable in providing navigation guidance associated with a route. A digital route marker may be referred to as a “digital navigation marker”, a “digital guidance marker”, a “digital navigation guidance marker” or the like. Various types of digital route marker are described herein, such as digital path markers, digital contextual markers and digital directional markers. A digital route marker may correspond to one or more positions in 3D space. A position in 3D space may be represented by one or more coordinates. Examples of coordinates include longitude, latitude and altitude. Any references made herein to defining or placing a digital route marker along a path, guiding a path-following person to a digital route marker, or the like should be understood to be referring to guiding to one or more positions in 3D space represented by the digital route marker. A digital route marker may be considered to be any digital description of location. A digital route marker may also describe orientation. One type of digital route marker is an AR anchor. An AR anchor may use GNSS coordinates. An AR anchor may use combinations of location descriptors including, but not limited to, IMU data to capture orientation. The term “digital path marker” is used herein to mean a digital representation of a position in 3D space associated with a path. References herein to a digital path marker representing a position in 3D space include the digital path marker representing multiple positions in 3D space unless the context implies otherwise. A digital path marker may take various different forms. For example, a digital path marker may be continuous or may be discrete. In some examples, a single digital path marker may be used to represent a path. For example, a single discrete digital path marker representing a position in 3D space corresponding to a given destination may represent a path from a current position to the given destination. In another example, a single, continuous digital path marker may represent a complete path to a given destination from a current position. In some examples, multiple digital path markers may be used to represent a path. The term “sequence” is used herein to mean an ordered set. The term “3D space” is used herein to mean any three-dimensional, real-world environment. The term “wearable” is used herein to mean anything that can be worn by a person. Examples of wearable devices include, but are not limited to, glasses, belts, shoes, vests, watches, earphones, headphones, and so on. The term “adjusting” is used herein to mean making any change. For example, references to adjusting route data or path data mean making any change to the route data or path data. Terms such as “changing”, “updating”, “modifying”, “adapting”, and the like may be used similarly herein. As such, references to “adjusting” an existing path include replacing the existing path with a new path, using the new path in addition to the existing path, merging the new and existing paths together, and so on. In some examples, adjusting an existing path may result in a new path having a different destination from a destination of the existing path. For example, an obstacle along the existing path may prevent the destination of the existing path being reached. An alternative destination may, however, be used in such situations. An alert may be issued to inform of the destination change. The term “obstacle” is used herein to mean anything that at least partly impedes walking along at least part of a path. An obstacle may be anything within a predetermined distance from a path that impedes walking movement. An obstacle may block a path. An obstacle may not physically block a path but may otherwise impede walking along the path. For example, ice may not physically block a path but may nevertheless impede walking along the path. As such, an obstacle may not physically prevent walking along a path but may make walking along a path more difficult, dangerous, or the like. An obstacle may not have been encountered when a path was defined. However, the obstacle may be encountered when the path is being followed. An obstacle may be a “dynamic” obstacle. A “dynamic” obstacle may be referred to as a “temporary” or “transient” obstacle. An example of a dynamic obstacle is a pedestrian. The pedestrian may not have been present along the path when the path was defined by the path-defining person, but may be present when the path is being followed by a path-following person. An obstacle may be a “permanent” obstacle. An example of a permanent obstacle is a building that has been constructed along a path since the path was defined. A given object may have been present when a path was defined, but may not have been an obstacle to the path-defining person. For example, the path-defining person may be of a height such that an overhanging branch was not an obstacle to them. However, the overhanging branch may be an obstacle to a taller path-following person following the path. The term “obstacle avoidance” is used herein to mean avoiding an obstacle in any manner. For example, both waiting for an obstacle to no longer be an obstacle, and guiding around an obstacle are types of obstacle avoidance as used herein. As such, when following a path, an obstacle can be detected, and walking trajectory may be adjusted. This can facilitate avoidance of temporary obstacles, such as other pedestrians. Obstacle avoidance may be performed by an obstacle avoidance system, which detects nearby obstacles. An obstacle avoidance system may use one or more sensors and process the sensor data generated by the sensor(s). Examples of sensors that may be used for obstacle avoidance include, but are not limited to, a camera, an ultrasound sensor and a LIDAR sensor. The term “digital contextual marker” is used herein to mean a digital representation of contextual information. A digital contextual marker may represent one or more positions in 3D space. The contextual information may be in a category, such as “warning” or “button”, and / or may include audio data such as a voice note. A digital contextual marker may also be referred to as a “digital context marker”. The term “contextual data” is used herein to mean any data associated with a digital contextual marker. “Contextual data” may be referred to herein as “contextual marker data”, “digital contextual marker data” or the like. The term “digital directional marker” is used herein to mean a digital representation of direction. A digital contextual marker may represent one or more positions in 3D space. A digital directional marker may also be referred to as a “digital direction marker”. The term “directional data” is used herein to mean any data associated with a digital directional marker. “Directional data” may be referred to herein as “directional marker data”, “digital directional marker data” or the like. The term “illuminate” is used herein to mean light up. In some examples, a navigation guidance device is operable to illuminaste some or all of a path. This may help alert nearby people that a user of the navigation guidance device is following the path. This may, for example, prompt such people to avoid walking along the path. Illuminating the path may improve operation of a navigation guidance device, for example in terms of computer vision effectiveness. A navigation guidance device may be operable in an illumination mode, in which some or all of a path is illuminated. A navigation guidance device may be operable in a non-illumination mode, in which some or all of a path is not illuminated. The navigation guidance device may enable a user to select between the illumination and non-illumination modes. The navigation guidance device may select between the illumination and non-illumination modes itself. The term “map-matching” is used herein to mean aligning real-world sensor data with a model of the world. For example, map-matching may place an icon indicating a location of a user on a pavement rather than in a road, where a pure GNSS location may position the user in the road. Map- matching may be used to “snap” GPS coordinates to roads based on a pre-existing map and knowledge that a vehicle is unlikely to have deviated off-road, for example into a field. Further understanding, such as assumptions of direction, may be used to infer which side of a road the user is on. The term “a” is used herein to encompass both the singular and plural, unless the context requires otherwise. Various examples will now be described with reference specifically to Figures 17A to 31 which relate to defining and / or following a path. Referring to Figures 17A to 17D, there is shown, schematically, an example of how a path may be defined. Examples are provided below which relate to how contextual information and / or directionality information associated with a path and / or route may also be defined. With reference to Figure 17A, a path is to be defined between an origin 164 and a destination 148. An origin may be referred to as a “start”, “starting point” or the like. A destination may be referred to as an “end”, an “end point” or the like. With reference to Figure 17B, a path-defining person 166 starts path definition at the origin 164. In this example, a first digital path marker 168 represents a position in 3D space corresponding to the origin 164. However, a digital path marker 168 may not be defined at a position in 3D space corresponding to the origin 164 in other examples. Definition of digital path markers 168 will be described in more detail below. With reference to Figure 17C, the path-defining person 166 is walking, or has walked, towards the destination 148 from the origin 164. In this example, a sequence of digital path markers 168 representing a sequence of corresponding positions in 3D space between the origin 164 and the current location of the path-defining person 166 has been defined. The sequence of digital path markers 168 digitally represents a path 170 between the origin 164 and the current location of the path-defining person 166 and a partial path between the origin 164 and the destination 148. With reference to Figure 17D, the path-defining person 166 has walked to, and reached, the destination 148. In this example, a sequence of digital path markers 168 representing a sequence of corresponding positions in 3D space between the origin 164 and the destination 148 has been defined. The sequence of digital path markers 168 represents a complete path 170 between the origin 164 and the destination 148. In this example, the path 170 is a straight-line path. However, a path can take a different form in other examples. As explained above, multiple different paths may exist between a given origin and destination. The path 170 may be referred to as a “defined path”, “existing path”, “already defined path”, “previously defined path” or the like. This is on the basis that the path 170 has now been defined by the path-defining person 166. Referring to Figures 18A to 18C, there is shown, schematically, an example of how at least part of the defined path 170 may be followed. With reference to Figure 18A, a path-following person 102 starts at the origin 164 and is guided towards a position in 3D space corresponding to a digital path marker 168 that follows a digital path marker 168 at the origin 164. In this example, the digital path marker 168 that follows the digital path marker 168 at the origin 164 is the next digital path marker 168 in the sequence of digital path markers 168 after the digital path marker 168 at the origin 164. With reference to Figure 18B, the path-following person 102 has been guided, sequentially, to several positions in 3D space corresponding to respective digital path markers 168 in the sequence of digital path markers 168 and is partway towards the destination 148. The next digital path marker 168 in the sequence of digital path markers 168 has been made available to the path-following person 102. Such sequential guidance enables the path-following person 102 to be guided in a desired direction relative to their current position. Without such sequential guidance, the path-following person 102 could, in principle, be located at a position in 3D space corresponding to one digital path marker 168 and could be equidistant from both a position in 3D space corresponding the digital path marker 168 that is next-closest to the origin 164 and a position in 3D space corresponding to the digital path marker 168 that is next-closest to the destination 148. If the path-following person 102 were to be guided to the position in 3D space corresponding to the digital path marker 168 that is next-closest to the origin 164, they would, in effect, retrace a portion of the path 170 along which they have already walked. This may not be effective where the path-following person 102 is seeking to reach the destination 148 efficiently. However, in some scenarios, it may be effective for the path-following person 102 to return to a position in 3D space corresponding to a previous digital path marker 168. For example, if an obstacle is detected along the path 170, the path-following person 102 may be guided back along a portion of the path 170, along which they have already walked. It is likely to be safe for the path- following person 102 to walk back along that already-walked portion of the path 170 to avoid the obstacle. With reference to Figure 18C, the path-following person 102 has reached the destination 148 by following the path 170 represented by the sequence of digital path markers 168. As such, the path-following person 102 has followed the path walked and defined by the path- defining person 166. In this example, the path-following person 102 is different from the path-defining person 166. However, the path-following person 102 may be the same as the path-defining person 166 in other examples. Referring to Figures 19A to 19D, there is shown, schematically, an example of how a path 170 may be adjusted. With reference to Figure 19A, the path-following person 102 is at the origin 164. The full existing, defined path 170 between the origin 164 and the destination 148 is shown in Figure 19A for ease of understanding. In this instance, however, an obstacle 150 is now blocking the path 170. However, the path-following person 102 may be unaware of the obstacle 150 while they are located at the origin 164. With reference to Figure 19B, the path-following person 102 has been guided, and has walked, to several positions in 3D space corresponding to several of the digital path markers 168 and is approaching the obstacle 150. In this example, a user device (not shown) of the path-following person 102 has detected the obstacle 150. More details on obstacle detection are provided below. The user device (not shown) determines that the path 170 should be adjusted based on the obstacle 150 and guides the path-following person 102 around the obstacle. With reference to Figure 19C, the path-following person 102 is now defining an adjusted path 172 and may be considered to be a path-defining person 166 as well as a path-following person 102. For ease of understanding, both the part of the defined path 170 the path-following person 102 has followed and the new path 172 that the path-following person 102 is defining are indicated in Figure 19C using reference sign “172” since the overall path the path-following person 102 walks along is different from the previously defined path 170. The size of the obstacle 150 has been exaggerated in Figure 19C for ease of understanding. In practice, an obstacle may be smaller relative to the overall path. With reference to Figure 19D, the path-following person 102 has reached the destination 148, having re-joined the previously defined path 170 after having been guided around the obstacle 150. In this example, the obstacle 150 is a permanent obstacle, such as a building that has been built since the previously defined path 170 was created. Path data representing the adjusted path 172 may be stored for future use. Path data representing the adjusted path 172 may replace path data representing the previously defined path 170, may be stored together with the path data representing the previously defined path 170, may be merged with the path data representing the previously defined path 170, or otherwise. The adjusted path 172 and the previously defined path 170 may therefore both be associated with the same route. Some or all of the path data representing the adjusted path 172 may be stored in route data associated with the route, potentially along with some or all of the path data representing the previously defined path 170. If the path-following person 102 and / or another path-following person were to request a path from the origin 164 to the destination 148, the adjusted path 172, avoiding the (permanent) obstacle, may therefore be provided. System Overview Referring to Figure 20, there is shown, there is shown an example of a system 400 in which a path may be defined and / or followed. In this example, the system 400 comprises a user device 174 of the path-defining person 166. The user device 174 may take various different forms, such as a smartphone, tablet computing device, sensory substitution device, and so on. In this example, the system comprises the user device 100 of the path-following person 102. In this example, the user devices 100, 174 are communicatively coupled to a server 176 via a communications network 178. The communications network 178 may, for example, comprise the Internet. In this example, the server 176 is communicatively coupled to a data store 180. Although the server 176 and data store 180 are depicted as separate entities in Figure 20, the distinction may be logical and the server 176 may comprise the data store 180. The system 400 may comprise different elements in other examples and is by way of a specific example only. The user device 100, the user device 174, and the server 176 may be considered to be “navigation guidance devices” in that they are used directly, or indirectly, to provide navigation guidance. Path Definition And Path Following Overview Referring to Figures 21A and 21B, there is shown an example of a method 500 in which a path is defined and followed. At item S21a, the user device 174 of the path-defining person 166 performs, or at least participates in, path definition. Path definition may be performed (including being initiated and completed) by the path-defining person 166 interacting with the user device 174 in a predetermined manner, or otherwise. At item S21b, path definition has been completed and the user device 174 sends path data to the server 176. The user device 174 may store the path data locally. At item S21c, the server 176 stores the path data in the data store 180. The server 176 may analyse, review, audit, or otherwise process the path data prior to storing the path data in the data store 180. At item S21d, the user device 100 of the path-following person 102 determines that the path- following person 102 wishes to follow the path defined by the path-defining person 166. At item S21e, the user device 100 of the path-following person 102 requests path data from the server 176 accordingly. To request the path data, the user device 100 may indicate the current location of the user device 100 and / or the origin and / or the intended destination and / or a path identifier (ID) and / or a route ID and / or any other information useable by the server 176 to identify the path data. At items S21f to S21h, the server 176 retrieves the path data from the data store 180 and sends the retrieved path data to the user device 100 of the path-following person 102. The user device 100 of the path-following person 102 may obtain the path data in a different manner in other examples. For example, the server 176 may intermittently provide some or all of the path data in the data store 180 to the user device 100. The user device 100 of the path-following person 102 may obtain the path data before the path-following person 102 desires to follow a particular path. For example, the user device 100 of the path-following person 102 may obtain path data relating to an underground path before the path-following person 102 walks along the underground path where the ability to obtain the path data may be limited, for example in terms of signal reception. The user device 100 of the path-following person 102 may have obtained the path data by generating the path data in another example. At item S21i, the user device 100 guides the path-following person 102 along at least part of a path represented by the path data. At item S21j, in this example, the user device 100 adjusts the path to be taken by the path- following person 102, for example as a result of detecting an obstacle along the existing path or otherwise. In this example, at items S21k and S21I, the user device 100 sends adjusted path data to the server 176 and the server 176 stores the adjusted path data in the data store 180. The server 176 may analyse, review, audit, or otherwise process the adjusted path data prior to storing the adjusted path data in the data store 180. At least some actions depicted by way of example only in Figures 21A and 21B may be performed in a different order, may be combined, may be separated, may be replaced, may be omitted, or may otherwise be varied. Although not shown in Figures 21A and 21B other messages, such as acknowledgment messages, may be sent. Path Definition Further examples concerning path definition will now be described. Referring to Figure 22, there is shown, schematically, an example of how a path may be defined. More specifically, Figure 22 shows an example of how a path-defining person 166 defines a path 170, by defining both a sequence of digital path markers 168 and a digital contextual marker 182. In this example, the path-defining person 166 holds the user device 174 in their right hand at the start of the path 170. In this example, the path-defining person 166 initiates path definition by issuing a voice command, such as “start recording”, to the user device 174. However, path definition may be initiated in a different manner in other examples. The path-defining person 166 walks along the path 170, with a sequence of digital path markers 168 being defined as the path-defining person 166 walks. In this example, each digital path marker 168 represents a position in 3D space at which the user device 174 is or was located. In this example, the destination of the path is a door 184. In this example, a final digital path marker 168 in the sequence of digital path markers 168 represents a position in 3D space at which the path-defining person 166 can stand to open the door 184. In this example, a digital contextual marker 182 is defined which represents a position in 3D space of a door handle 186 of the door 184. Digital contextual markers 182 will be described in more detail below. In this example, the user device 174 has obtained, by generating, path data that represents the sequence of digital path markers 168 and contextual data that represents the digital contextual marker 182. As can be seen, schematically, in Figure 22, the hand in which the path-defining person 166 holds the user device 174 may affect the positions in 3D space represented by the digital path markers 168 relative to the centreline of the path-defining person 166. For example, if the path-defining person 166 had held the user device 174 in their left hand instead of their right hand, the positions in 3D space represented by the digital path markers 168 may have been to the left of the centreline of the path-defining person 166 rather than to the right. This depends, to some degree, on the level of accuracy at which the location of the user device 174 can be determined and / or at which the positions of the digital path markers 168 can be defined. Further details on this, and how handedness compensation may be applied, are provided below. Referring to Figure 23, there is shown, schematically, another example of how a path may be defined. More specifically, Figure 23 shows an example of a path-defining person 166 being guided by another person 166 while defining the path. A full path 170 as will be defined is shown in Figure 23 for ease of understanding. However, it should be understood that the full path 170 has not yet been defined in the scenario shown in Figure 23. In this example, the other person 166 is shown in Figure 23 with reference sign “166” since, in this example, the path-defining person 166 and the other person 166 are both path-defining. In this example, the other person 166 guides the path-defining person 166 along the path, and the user device 100 of the path-defining person 166 defines a sequence of digital path markers 168 representing positions in 3D space along the path 170. In another example, a guide dog may guide the path-defining person 166. Additionally, in this example, the other person 166 guides the path-defining person 166 such that the path-defining person 166 avoids multiple obstacles 150a, 150b. In this example, one obstacle 150a is a wheelie bin and the other obstacle is a lamppost 150b. However, the obstacles 150a, 150b may be different in other examples. In this example, the other person 166 guides the path-defining person 166 along a pavement 188 and prevents the path-defining person 166 from walking on a road 190. This increases safety for both the path-defining person 166 and the other person 166 while defining the path, and also defines a safe path, avoiding the road 190, for path-following people to follow. One or more digital contextual markers may be defined to indicate the presence of the road 190. The digital contextual marker(s) may warn path-following people of the road 190 that is near to the path 170. The one or more digital contextual markers may represent a plane in 3D space which runs vertically along an edge of the pavement. Obstacle Detection Obstacles may be detected in various different ways. For example, if the user device comprises a sufficiently high-quality camera, computer vision may be used. LIDAR may be used. However, while LIDAR provides depth information, LIDAR may not provide particularly high-resolution output and so may be used conservatively and / or may provide limited obstacle detection functionality. LIDAR is available on some current smartphones. However, LIDAR is typically used on current smartphones to focus a camera. LIDAR may also be used on current smartphones for object-scanning, mapping, and in AR applications to assign surfaces. However, in accordance with examples described herein, LIDAR may be used for path-planning and / or obstacle detection and / or obstacle avoidance. Path-Following — Obstacle Avoidance Further examples concerning path-following will now be described. In some examples, a path is adjusted in response to a path adjustment trigger. Various example path adjustment triggers are described herein. Examples of path adjustment triggers include, but are not limited to, obstacle detection, non-obstacle detection, expiry of a timer and a user- initiated request to adjust a path. Referring to Figure 24, there is shown, schematically, an example of how a path may be followed and adjusted to avoid an obstacle. More specifically, Figure 24 shows an example of how a path-following user 102 initially follows a defined path, but adjusts the path to avoid an obstacle 105a that has moved from not being along the defined path to being along the defined path. Referring also to Figure 23, the wheelie bin obstacle 150a has moved from an original position (indicated in Figure 24 by a broken line box) to a new position, which is along the defined path. In this example, the position in 3D space represented by one of the digital path markers 168 of the defined path coincides with the location of the wheelie bin obstacle 150a in 3D space. However, an obstacle may not coincide with a position in 3D space represented by digital path marker 168 in other examples. For example, an obstacle may be between positions in 3D space represented by digital path markers 168 in other examples. In this example, the user device 100 comprises a spatial sensor and uses spatial sensing to sense the wheelie bin obstacle 150a as the path-following user 102 initially follows the defined path. In response to sensing the wheelie bin obstacle 150a, the user device 100 adjusts the path along which the path-following user 102 is to be guided, and guides the path-following user 102 around the wheelie bin obstacle 150a. As such, in this example, the user device 100 has performed obstacle avoidance path adjustment in respect of the wheelie bin obstacle 150a. In this example, the adjusted path 172 converges with the defined path after the path- following user 102 has walked around the wheelie bin obstacle 150a. However, paths may not converge in this manner in other examples. In this example, the lamppost obstacle 150b is a permanent obstacle and the path does not need to be adjusted to avoid the lamppost obstacle 150b in this instance. This is because, in this example, the lamppost obstacle 150b was not along the defined path and is not along the adjusted path. Various actions may be taken based on an adjusted path. In this specific example, it may be determined that path data representing the adjusted path 172 should not overwrite path data representing the defined path. For example, since the wheelie bin obstacle 150a is a temporary obstacle, it may no longer be an obstacle when future path-following people follow the previously defined path. In another example, it may be determined that path data representing the adjusted path 172 should be stored in addition to data representing the defined path. For example, the wheelie bin obstacle 150a may be in the location shown in Figure 24 only for a certain number of hours on a certain day of the week, such as when the wheelie bin obstacle 150a is to be emptied, with the wheelie bin obstacle 150a being in the location shown in Figure 23 at other times. In such an example, the adjusted path may be used at times of day and on days of the week when the wheelie bin obstacle 150a is to be emptied and the defined path may be used at other times. In this example, digital path markers 168 are defined after the adjusted route has been walked along. However, digital path markers 168 may alternatively or additionally be defined before the adjusted route has been walked along and the path-following user 102 can be guided based on the digital path markers 168 accordingly. Referring to Figure 25, there is shown, schematically, another example of how a path may be followed and adjusted to avoid an obstacle 150. More specifically, Figure 25 shows a path-following person 102 following a path and where obstacle avoidance adjustments are made to the path. In this example, the path-following person 102 is following a defined path and encounters an obstacle 150 in the form of another pedestrian. In this example, the user device 100 guides the path- following person 102 around the other pedestrian 150, adjusting the path 172 the path-following person 102 is following accordingly. In this example, the path-following person 102 re-joins the defined path after having navigated around the other pedestrian 150. Although, in this example, the user device 100 guides the path-following person 102 around the other pedestrian 150, one or more other avoidance actions could be taken in other examples. For example, the user device 100 may recognise that the obstacle 150 is a pedestrian and may prompt the path-following person 102 to stand still temporarily in the expectation that the pedestrian will continue their journey shortly and will no longer be an obstacle. The user device 100 may request the pedestrian to move such that they are no longer blocking the defined path. The user device 100 may predict movement of the pedestrian and determine that the pedestrian will no longer be blocking the defined path when the path-following person 102 arrives at the current location of the pedestrian and may not adjust the path accordingly. Such prediction may be based on the speed and / or direction of movement of the pedestrian, the current location of the pedestrian, or otherwise. As such, various different avoidance actions may be used. Path-Following — Non-Obstacle Avoidance And / Or Adjustment In some examples, a path is adjusted to avoid something other than an obstacle, in other words a “non-obstacle”. For example, a path may have been defined in good weather which includes indoor and outdoor path portions. If the path is being followed when it is raining outside, at least the outdoor path portion may be adjusted to avoid the rain outdoors. For example, there may be a less efficient path that is fully indoors. Although the path is less efficient, it may be selected to keep the path-following person dry, may be safer than the outside path portion, and so on. In some examples, a path may be adjusted even if there is nothing along the path to avoid. For example, a path-following person may indicate that they wish to follow a longer path for increased exercise. Digital Contextual Markers As explained above, a digital contextual marker may be defined along, or otherwise in association with, a path. References to defining a digital contextual marker “along” a path should be understood to encompass any definition such that a path-following person may encounter the digital contextual marker as they walk along the path. In some examples, a digital contextual marker is defined in addition to one or more digital path markers. A digital contextual marker is a type of digital route marker with an associated context. The term “context” Is used herein to mean contextual information associated with an environment in which a person is located. Examples of such contexts include, but are not limited to, warnings, dangers, points of interest, key objects and key locations. References to a digital contextual marker having an associated context encompass the digital contextual marker having multiple associated contexts. For example, a digital contextual marker may be associated with a rotating entrance door to a building (a key object) while also being associated with a “rotating door” warning. A digital contextual marker may have an activation area. An activation area may be referred to herein as an “area of effect”, “area of operation”, “context area”, “active area”, “trigger area” or the like. When the user device is within the activation area of the digital contextual marker, the digital contextual marker is activated. The activation area may take various different forms. For example, the activation area may be of a regular or irregular shape. An activation area may be a 1D, 2D or 3D area. The activation area of a digital contextual marker may be dependent on one or more activation area parameters. An example of an activation area parameter is a level of danger associated with the digital contextual marker. Different digital contextual markers may have different activation areas from each other, for example where multiple digital contextual markers are defined along a path. For example, a digital contextual marker associated with a high level of danger may have a larger activation area than an activation area of a digital contextual marker associated with a low level of danger. As such, the activation area of a digital contextual marker may depend on a purpose of the digital contextual marker. In particular, an activation area may be wide or may be very precise depending on the purpose of the digital contextual marker. By way of specific, non-limiting examples, a digital contextual marker having a large activation area may be used for a road crossing, a building site, a busy driveway, or something else that may pose a hazard. By way of a specific, non-limiting example, a digital contextual marker having a small activation area may be used for a precise key object. Examples of such key objects include, but are not limited to, a button for crossing a road and a button for calling an elevator (also known as a “lift”). An object and / or a digital contextual marker may be associated with multiple activation areas. Different activation areas may be associated with different contexts. For example, a handrail for a set of stairs may be associated with a wide activation area in connection with a hazard context and a more precise activation area in connection with a key object context. A digital contextual marker may be associated with audio information. Such audio information may be output whenever the digital contextual marker is activated. Alternatively, such audio information may be made available to be output whenever a digital contextual marker is activated, but predetermined user action may then trigger the audio information to be output. Examples of audio information include, but are not limited to, voice recordings and voice notes. Such audio information may be used, for example, to denote which shops are nearby. Such audio information may be used in a leisure sense, for example to perform a guided tour. Such a guided tour may be of a museum or of a street along which a path-following person is walking. A digital contextual marker may be associated with haptic information. Such haptic information may be output whenever the digital contextual marker is activated. The haptic information may, for example, represent a specific vibration pattern to be output when the digital contextual marker is activated. A digital contextual marker may provide any type of context. For example, for an automatic door, a digital contextual marker may indicate that the door is automatic, where a path-following person should stop to trigger the automatic door and / or where a path-following person should stand so as not to come into contact with the automatic door when the automatic door opens. Path-adjustment may be dependent on one or more digital contextual markers. For example, if a digital contextual marker indicates a highly dangerous area, path adjustment may be restricted to small deviations from a walked path in the vicinity, such as in the activation area, of the digital contextual marker in view of the local level of danger. As such, a digital contextual marker may be used to influence obstacle-avoidance-related path changes. If an area in which an obstacle is located is highly dangerous, then very little path deviation may be allowed despite the obstacle. A digital contextual marker may be used in conjunction with preference data of a path- following person. For example, preference data of a given path-following person may specify that the given path-following person should not be made aware of certain types of digital contextual marker. Referring to Figure 26, there is shown, schematically, an example of how a digital contextual marker may be activated and used. More specifically, Figure 26 shows an example of how a digital contextual marker 182 may be used at a street crossing. Figure 26 also shows how a ranged alert 192 may be issued and how the path-following person 102 may be guided precisely to the digital contextual marker 182. In this example, the path-following person 102 is following a path towards a lamppost 194 with a crossing button 196. Contextual markers can relate to multiple purposes and can trigger multiple communications from the user device 100 to the path-following person 102. The digital contextual marker 182 has been defined at a position in 3D space coinciding with the crossing button 196 of the lamppost 194. In this example, the digital contextual marker 182 first has a relatively large activation area and the path-following person 102 is provided with the ranged alert 192 as they approach the street crossing and are within the large activation area of the digital contextual marker 182. The ranged alert 192 may take various different forms including, but not limited to, audio and / or vibration. As the path-following person 102 reaches the lamppost 194, the path-following person 102 is guided towards a precise position in 3D space represented by the digital contextual marker 182 and, hence, to the crossing button 196 itself. The wide area alert during the approach to the crossing may include information about upcoming precision guidance to the crossing button 196 or it may be disabled by users preferentially. Directionality In some examples, directional data is used. For example, a digital path marker 168 and / or a digital contextual marker 182 may comprise, or otherwise be associated with, directional data. In some examples, a dedicated digital directional marker is provided. The term “digital directional marker” is used herein to encompass any digital route marker that is associated with directional data; whether a dedicated digital directional marker or otherwise. Referring to Figure 27, there is shown schematically, an example of how directional data may be used. More specifically, Figure 27 shows an example of how a digital directional marker may be used to approach a bench 198. In this example, the path-following person 102 approaches a bench 198 with their user device 100 and reaches a digital path marker 168. In this example, the digital path marker 168 comprises, or is otherwise associated with, a digital contextual marker 182. In this example, the digital contextual marker 182 comprises, or is otherwise associated with, directional data that indicates a direction, or orientation, of approach such that the path-following person 102 approaches the seat side 1100 of the bench 198. This enables the path-following person 102 to approach the bench 198 from the correct side. With the specific example of the bench 198, the path-following person 102 will know to leave the bench 198 in the opposite direction to the direction of approach. However, directional data may indicate a direction of approach and / or departure in other examples. By way of another example, door openings and door handles may also be direction- dependent. A path may include a door handle on the same side of the door as a path-following person as well as on the other side of the door. The path-following person should be directed to the correct door handle, based on the path along which they are walking, and not to the wrong door handle. In some examples, the directional data may indicate the direction in which a door is going to open. The path-following person can thereby be alerted or otherwise informed accordingly. Manual Path Definition As explained above, a path-defining person may manually define digital path markers based on a walked path. For example, the path-defining person may carry a user device as they walk along the path and may interact with the user device whenever they wish to define a digital path marker representing a position in 3D space along the path. Such interaction may involve pressing a hard or soft button on the user device, issuing a voice command to define a digital path marker, moving the user device in a predetermined manner when a digital path marker is to be defined, or otherwise. Automated Path Definition As also explained above, digital path markers may be defined automatically. In other words, digital path markers may be defined without a path-defining person having to define each digital path marker manually. In such examples, the path-defining person still walks along a path, but digital path markers may be defined by an entity other than the path-defining person. For example, a path-defining person may have a user device that defines digital path markers without the path-defining person having to interact with the user device to define each such digital path marker. The user device may, for example, define digital path markers in accordance with one or more digital path marker definition conditions. Examples of digital path marker definition conditions include, but are not limited to, definition every predetermined amount of time, definition every predetermined distance walked, definition when a walking direction deviates from a previous walking direction by more than a threshold amount, and so on. In some examples, however, there may be minimal interaction with the user device to enable the digital path markers to be defined automatically. For example, the path-defining person may give permission for the user device to define digital path markers automatically the first time or each time the path-defining person defines a path, but the path-defining person may not need to interact with the user device each time a digital path marker is defined. As such, simply having a tracker user device in a pocket may be enough to enable a path- defining person to define a path based on automatic digital path marker definition and, potentially, map-matching. Automated path definition may also be effective where the path-defining person does not possess user equipment to be able to define the digital path markers, but can otherwise assist in defining the path. For example, a transport station may have a series of video cameras that can track a path walked by a given path-defining person through the transport station even if the path-defining person is not carrying a user device. One or more digital path markers may be defined based on the walked path using such tracking. Digital Path Marker Establishment In some examples, a user device is able to cause a digital path marker to be stored in a centralised data store of digital path markers and thereby establish the digital path marker. In such examples, the user device may be authoritative in that the user device can authoritatively establish the digital path marker. In other examples, a different authoritative entity may decide whether or not a suggested digital path marker should be established. Such an authoritative entity may, for example, verify that the suggested digital path marker represents in a safe position in 3D space. Digital Contextual Marker Definition As with digital path marker definition, digital contextual marker definition may be manual or automated. In particular, a digital contextual marker may be defined based on activity of the path-defining person and / or based on activity of one or more other people. Such activity may include behaviour of the person(s) concerned. For example, if a path-defining person moves upwards at an angle but is not walking, a “moving staircase” digital contextual marker may be suggested and / or defined automatically. By way of another example, a user device may be able to detect loud road noise at a given location, for example using a microphone of the user device. Such road noise may indicate a busy and / or dangerous junction. An appropriate warning digital contextual marker may be defined automatically at a position in 3D space corresponding to the given location on this basis. Path Adjustments — Deviations If one or more path-following users deviate from a path by more than a predetermined amount, path data may be adjusted to include and / or account for such deviation. The predetermined amount may comprise a threshold number of deviations. For example, if the path is deviated from more than ten times, the path data may be adjusted. The predetermined amount may comprise a threshold distance. For example, deviations may be thresholded such that small deviations associated with natural human movement may be smoothed out or ignored. Such adjustment may be as a result of a permanent or temporary obstacle. Such adjustment may itself be permanent or temporary. For example, if a new permanent obstacle, such as a bench, is installed in a street, then a permanent adjustment may be made. Some obstacles are, or at least may be, temporary. For example, building works and road works are likely to be temporary. Obstacle avoidance may be used to check whether a given obstacle that was along a previous path is no longer present, such that the previous path may be reverted to and opened again. This may take place even though a user is being guided along an updated path that avoids the obstacle. In particular, the previous path may be the most efficient path between an origin and a destination in the absence of the obstacle, and so may be reverted to once the obstacle is no longer along the previous path. In some examples, a timer is set in association with a given route and / or path. If it is known that an obstacle along a previous path will no longer be present at a particular time, then the timer can be set to expire at that particular time, and the previous path may be reverted to on expiry of the timer. As such, where a previous path is known to be open again on a set date, the previous path may be attempted again following reopening of the previous path. In some examples, a path-following person may be alerted in connection with detection of an obstacle. For example, when obstacle avoidance causes significant path adjustment and / or when an obstacle blocks a path, the path-following person may be alerted. The alert may indicate one or more obstacle properties of the obstacle. Examples of obstacle properties include, but are not limited to, size, type, and movement. Such an alert can help the path-following person understand the nature of the obstacle. In particular, the path-following person may exercise more caution if the obstacle is particularly dangerous. For example, the path-following person may walk more carefully along any alternative path if ice is detected along an existing path. In some examples, audio (for example, voice), image, video and / or other data is analysed to assess an obstacle. Such analysis may be automated, may be performed by a human moderator, or may use a hybrid analysis technique. For example, a user device might incorrectly recognise a shadow on the floor as an obstacle in the form of a hole in the floor. Further analysis may identify that the perceived obstacle is not, in fact, an obstacle. Obstacle Data Obstacle data representing one or more known obstacles may be generated, stored, or otherwise used. Obstacle data may, for example, indicate one or more obstacle attributes of one or more known obstacles. Examples of obstacle attributes include, but are not limited to, size, position in 3D space and type. Existing paths may be analysed to determine whether known obstacles are, or may be, an obstacle along another path. For example, on detection of a new obstacle, existing paths that are nearby the obstacle may be analysed. Existing paths which may have a nearby obstacle may, for example, be prioritised for further review, may be given a lower priority than other routes, or otherwise. Path Adjustments — Path Efficiency Path efficiency is a measure of the directness of a path when compared to an ideal path. In accordance with examples, path efficiency may be optimised. For example, a path may be smoothed such that the path gets incrementally straighter over time. Map-matching may be performed to estimate whether an ideal path is straight and / or where a path should bend. For example, the path may bend around a corner. This can be used to account for unnecessary or undesirable deviations from an ideal path that were made when the path was defined. For example, if a path through a high street was defined on a day when the high street was particularly busy, the initially defined path may have significant undulations and deviations from an ideal, straight path along the high street. Voluntary Stopping In some examples, a path-following person may not follow a path in an expected manner, or at least in the same manner as a path-defining person. For example, the path-following person may unexpectedly stop walking along the path to take a break. In particular, the path-following person might not stop walking because of an obstacle that may otherwise cause the path to be adjusted. A user device of the path-following person may detect when the path-following person is stationary. If the user device does not detect an obstacle along the path, then one or more predetermined actions may be taken. In such a situation, the path-following person may be considered to be stationary voluntarily and not as a result of an obstacle. An example of such a predetermined action is relaxing the user device to save battery. Another example is inhibiting a path adjustment feature from recording deviations from the path, for example while the path-following person finds a bench on which to sit to rest. Holding Offset A human being is approximately 0.5 metres wide. In some examples, a path is adjusted dependent on the side of the body on which a path- following person is holding the user device. For example, when guiding a path-following person through certain spaces, such as doorways, the path can be adjusted depending on whether the path- following person is holding the user device in their left or right hand. Defining a path directly down the middle of the free space in the doorway, to reduce the likelihood of the path-following person colliding with the doorway, would involve the path-following person holding the user device directly in front of themselves to avoid collision. While this is possible, it can be particularly tiring for a path- following person and does not necessarily represent natural behaviour. As explained above, obstacle avoidance may adjust a path. However, knowing or detecting the side of the body on which the path-following person is holding the user device provides improved guidance and path adjustment. Referring to Figure 28, there is shown, schematically, an example of how handedness may be used. More specifically, Figure 28 shows an example of a user-optimal path based on a location of the path-following person 102a, 102b relative to the user device 100a, 100b, using a doorway 1102a, 1102b as an example. In this example, a first scenario is shown in which a path-following person 102a approaches a doorway 1102a with the user device 100a in their right hand. The device-optimal path coincides with the centreline of free space through the doorway 1102a. However, even though the width of the path- following person 102a is less than the width of the doorway 110243, the path-following person 102a is off-centre with respect to the doorway 1102a and, in this example, collides with the doorway 1102a. The path-following person 102a is off-centre with respect to the doorway 1102a because the user device 100a is in the centre of the doorway 1102a and the centre of the path-following person 102a is to the left of the centre of the doorway 1102a in the direction in which the path-following person 102a is walking. In this example, a second scenario is shown in which the path through the doorway 1102b is offset relative to the centreline of free space. This provides a user-optimal path through the doorway 1102b such that the path-following person 102b does not collide with the doorway 1102b. When path data is created, for example when a path is defined and / or adjusted, handedness metadata may also be created. The handedness metadata may be used to determine whether, and if so what, offset should be applied when the path is followed. For example, if a path is initially defined by a right-handed person, handedness metadata may be stored with path data representing the path. If a left-handed path-following person follows that path in the future and if, for example, that path-following person is approaching a doorway, an appropriate offset can be applied to the initially defined path to account for the different handedness. The width of the path-following person may be taken into account in determining the offset. Smoothing may be applied such that an offset is gradually applied, for example as the path- following person a doorway. Path Choice In some examples, a path-following person is alerted when multiple paths are available. Such an alert may be in the form of an audio prompt or otherwise. The path-following person may receive such an alert before they start following a path and / or while they are following a path. The alert may indicate that the path-following person may take another, “branching” path. A branching path may be a sub-path. The sub-path may be short relative to the main path. For example, the main path may be from the home of a path-following person to a town centre, the path- following person may be following the path towards the town centre, and the branching path may be a sub-path from the current location of the path-following person to a door handle of a particular shop along the main path. In such examples, the path-following person may be expected to re-join the main path after they have completed the sub-path, i.e. visited the shop. A branching path may be an alternative path to a given destination or to a different destination. For example, on a hike, there may be an intersection (often referred to as a “fork”) in a road. The road may diverge into one road segment associated with the main path and another road segment associated with an alternative path. As such, a branching path may represent a larger detour than a sub-path. Multiple paths may be chained together. For example, a path-following person may request guidance to a given shop in a town centre. A routing engine may initially use a general “town centre” path to guide the user into the town centre, followed by a more specific path to guide the path- following person from a given town centre location to the door handle of the given shop. The routing engine may automatically change from the general path to the specific path. Such change may be automatic in that the path-following person may not be required to interact with their user device in real time to effect the path change. In some examples, multiple versions (or “recordings” of a given path may be defined and / or multiple different paths to a given destination (or “location”) may be defined. This enables a preferred version and / or path to be selected. Such selection may be automatic and / or may be by the path- following person. Different versions of a given path may, for example, comprise different numbers of digital path markers, different digital contextual markers, or otherwise. As such, in some examples, different path-following people may be provided with different versions and / or paths from the same starting location. The preferred version and / or path may be selected based on one or more path-selection attributes. An example of a path-selection attribute is a user-related path-selection attribute. The term “user-related path-selection attribute” is used herein to mean an attribute of and / or which affects a user. Examples of user-related path-selection attributes include, but are not limited to, lighting, time of year, time of day, weather conditions, fitness levels, and specific visual impairment. For example, a preferred version and / or path may avoid icy patches. In terms of the specific vision impairment of the user, most registered blind individuals have some residual sight. Another example of a path-selection attribute is a user-device-related path-selection attribute. The term “user-device-related path-selection attribute” is used herein to mean an attribute of and / or which affects a user device of a user. For example, a user-device-related path-selection attribute may include an attribute that impedes the ability of the user device to function. An example of a user-device-related path-selection attribute is an overhead structure blocking GPS signal reception. Some path-selection attributes may be both user-related path-selection attributes and user- device-related path-selection attributes. For example, lighting level can affect both the user and the user device. Multiple path-selection attributes may be used in some examples. In such examples, the path- selection attributes may be of the same or different types. In particular, multiple user-related path- selection attributes may be used together, such as lighting conditions and the specific vision impairment of the user. For example, the specific vision impairment of the user may be that they are very sensitive to light and so may prefer dark locations, such as dark tunnels, over bright locations. In other examples, users with another specific vision impairment may prefer large amounts of direct sunlight over darker locations. In some examples, path-selection metadata may be recorded when a path is defined. Examples of such path-selection metadata include, but are not limited to, camera data, lighting condition data, time of data recording, and day of week. Such path-selection metadata may be assessed to provide improved path recommendations to future path-following people for a given route and / or path, for example based on their sight. Walking direction may affect path choice. For example, one version of a given path may be useable for a path-following person walking from a first location to a second location and another version of the given path may be useable for a path-following person walking from the second location to the first location. This may be effective in, for example, guiding a path-following person away from trying to walk down a moving staircase that is going up. By way of another example, some train stations have separate entrances and exits, and a path-following person can be guided towards an entrance or exit depending on whether they are entering or leaving the train station. As explained above, a digital contextual marker may record such information, for example by manual notation when the path is initially defined. As such, a path may be modified or changed to select a preferred path. Referring to Figure 29, there is shown, schematically, an example of how path choice may be used. More specifically, Figure 29 shows an example of a path-following person 102 walking towards a town centre, with multiple path options and a path option alert 1104. In this example, the path-following person 102 is following a main path 170 into a town centre. As the path-following person 102 is walking along the main path 170, the user device 100 issues an alert 1104 to notify the path-following person 102 of an alternative path option. In this example, the alert 1104 comprises a voice alert, such as “path option to shop”. In this example, the alert 1104 also comprises a vibration alert. In this example, the alternative path option is a sub-path 1106, which leads the path-following person 102 to a position in 3D space corresponding to a door handle of a door of the shop as represented by a digital contextual marker 182. The path-following person 102 could visit the shop by following the sub-path 1106, could follow the sub-path 1106 in the reverse direction after having visited the shop, and could then re-join the main path 170 to continue along the main path 170 after the detour to the shop. Path Usage A path may be crowdsourced, open-sourced, and / or community-driven. A path may be used in various ways. For example, a path may be shared, ranked, copied and / or adjusted. In some examples, some or all such usage is restricted to the path-defining person that defined the path. In some examples, one or more people in addition to or instead of the path- defining person may be permitted to use such a path. As such, a defined path may be useable by a person that did not originally define the path. A path-following person navigating along a path may use the same hardware (for example, user device) as a path-defining person, or may use different hardware. In the latter case, the hardware may be significantly different. For example, the hardware used by the path-following person navigating along the path and / or the path-defining person may comprise a smartphone with a sensor attachment or may comprise a haptic device with its own sensor(s). The path-following person navigating along a path may or may not be vision impaired. For example, the path-following person may be an employee of a path-maintaining company, may be a friend or carer of a vision impaired person seeking navigation guidance in a dark environment, or otherwise. In addition, example navigation guidance devices described herein may be effective for people with dementia by providing particularly natural and intuitive navigation guidance. Referring to Figure 30, there is shown, schematically, an example of how path data may be used. More specifically, Figure 30 shows an example of how path data can be shared and / or ranked (also referred to herein as “rated”) between users. In this example, a user device 174 is used to define a path, which results in path data 1108 being generated. The path data 1108 comprises path data representing the defined path. The user device 174 uploads the path data 1108 to an online data store 180, for example via a server (not shown). Other user devices 102a, 102b can download the path data 1108 from the online data store 180, for example via a server (not shown). Users (not shown) of the other user devices 100a, 100b may rank the path data 1108, resulting in ranking data 1110a, 1110b associated with the corresponding path. The user devices 102a, 102b may upload the ranking data 1110a, 1110b to the online data store 180, for example via a server (not shown). Although not shown in Figure 30, route data, contextual data, directional data, and / or other data described herein may be used, for example shared and / or ranked, in a similar manner. Although, in this example, users (not shown) of the other user devices 100a, 100b rank the path data 1108, automated ranking may alternatively or additionally be used in other examples. Automated ranking may, for example, analyse an extent to which path-following people deviate from a given path. A relatively high ranking may be used where such deviation is relatively low, and a relatively low ranking may be used where such deviation is relatively high. 3D Obstacle Avoidance An obstacle is not necessarily in the same horizonal plane as a user device of a path-following person. In particular, an obstacle may be overhead or may be beneath such a plane. For example, a low branch in a park may be overhead and a hole in the ground may be beneath the plane of the user device of the path-following person. In examples, 3D obstacle avoidance can detect, and avoid, such obstacles. User Device Accessory In some examples, a user device accessory enables an existing user device to have at least some of the functionality described herein. For example, relatively few smartphones currently have LIDAR sensors. An accessory comprising a LIDAR sensor may be provided for coupling with such a smartphone such that the smartphone can use LIDAR sensor functionality. Such an accessory may be coupled with the smartphone using a wired connection or a wireless connection. Data Processing Division In some examples, a navigation guidance device comprises multiple component elements, such as shown in Figure 7. In such examples, some data used for navigation guidance may be processed by one of the elements and other data used for navigation guidance may be processed by another one of the elements. For example, referring to the example sensory substitution device 100 shown in Figure 7, the handheld body portion 130 may be used to process outdoor location information, which is typically less complicated to process than indoor location information. For example, the handheld body portion 130 may process LIDAR data. The smartphone portion 134 may be used to process other information, such as camera data, which is typically more complicated to process. In some examples, data used for navigation guidance is processed outside of the navigation guidance device. For example, object recognition may be cloud-based. As such, data processing may be divided based, for example, on the type of data. Example Implementation 1 By way of example only, a specific implementation example will now be provided. Using ARKit from Apple™, a carer may walk along a path to be followed by a vision impaired person. The carer may tap the screen of their user device to define digital path markers at regular intervals corresponding to positions in 3D space along the path. The carer may also tap the screen of their user device to define digital contextual markers representing positions in 3D space of key objects along the path, such as on door handles and handrails at stairs. The carer may also provide contextual information for the digital contextual markers. The vision impaired person, namely the path-following person, then loads the path defined by the carer onto a user device of the vision impaired person. The user may load the path using a voice command or otherwise. The user device of the vision impaired person detects at which end of the path the vision impaired person is located. The user device of the vision impaired person then proceeds to guide the vision impaired person to the other end of the path. In this example, the user device of the vision impaired person comprises a handheld, haptic device. When the vision impaired person approaches positions in 3D space represented by the key digital contextual markers, for example as determined by their respective activation areas, the vision impaired person is made aware of the key digital contextual markers. For example, the user device of the vision impaired person may generate an audio prompt. A key digital contextual marker may, for example, represent a handrail ahead of a set of stairs. When the vision impaired person navigates busy corridors, streets, or other environments, obstacle avoidance allows path adjustment by detecting and guiding the vision impaired person around obstacles. The vision impaired person may be informed, for example by audio output, of a path adjustment and / or path blockage. This may depend on one or more properties of the adjustment and / or path blockage. An example of such a property is a size of the deviation. For example, if the deviation is below a threshold value (which may, for example, correspond to 10 centimetres), the vision impaired person may not be informed. The vision impaired person may have previously specified that they do not wish to be informed of such small deviations, for example. If, however, the deviation is above the same or another threshold value (which may, for example, correspond to 1 metre) and / or if there is total blockage of a path, the vision impaired person may be informed. In some examples the vision impaired person is always informed of a large deviation and / or total path blockage. If the path appears to be untraversable in view of an obstacle, the vision impaired person can be informed accordingly. Additionally, or alternatively, sensor data may be analysed to attempt to discern the cause of the obstacle. For example, regularly moving LIDAR data may indicate a slow- moving queue, such as in a train station. By way of another example, another type of sensor, such as a camera, may be used to perform object detection and may identify, for example, the object as being a building site. A path may be publicly shared with other users, such as other vision impaired users. Other users may rank the path. In some examples, a user is only able to rank a path if they have followed some or all of that path. A path may, alternatively, be private. For example, a path to the home address of a user may be kept private. Example Implementation 2 By way of example only, another specific implementation example will now be provided. In this example, a vision impaired user holds a haptic device which comprises a high-precision (for example, centimetre-accuracy) inertial navigation system. The vision impaired user issues a voice command to start defining a new path. The vision impaired user is then guided by another person, such as a friend, along the path. While defining the path, the vision impaired user can use voice commands and / or a button to trigger their device to define digital contextual markers representing key positions in 3D space while being guided by the other person. This allows digital contextual markers to be defined for, for example, buttons at traffic lights, door handles of shops, and so on. This also allows warnings to be defined before large roads and / or other hazards. Once the vision impaired user has defined the path, the vision impaired user may then start following the path from a position in 3D space represented any digital path marker associated with the path. If the vision impaired user is not close to the path, then they are guided to a position in 3D space represented by given digital path marker associated with the path. The position in 3D space to which the vision impaired user is directed may be the closest position in 3D space to the user device of the vision impaired user, or otherwise. If the closest position in 3D space is more than a threshold distance away, for example at least two metres through an untested area, the vision impaired user is warned. Moving Environments The frame of reference of a path may move, such as when the path is at least partly located on a bus, train, aircraft or boat. A local detector of a user device can determine whether a user is moving, or whether an environment in which the user is located is moving. For example, data from a local spatial sensor, which may indicate that a user is stationary may conflict with data from a localisation sensor, such as GPS, which may indicate that the user is moving at 40 kilometres per hour. Moving environment determination may also be performed using detection of walking movements, for example based on an accelerometer or IMU. When defining a path, detection of a moving environment event may enable a path to be split into multiple smaller paths and then linked together. For example, a path-defining person may be defining a path from their home to their work. The path may involve the path-defining person walking from their home to a first bus stop, getting a bus to a second bus stop, and then walking from the second bus stop to their work. The path may be split into a smaller path from their home to the first bus stop and another smaller path from the second bus stop to their work. The home-to-first-bus-stop path may be used on its own or in a chain to navigate to the work location. Route Data Referring to Figure 31, there is shown, schematically, an example of how route data may be arranged. In this example, the route data 1112 is associated with a given route. In this example, the route data 1112 comprises a route ID 1114. The route ID 1114 may uniquely identify the given route amongst other routes. In this example, the route data 1112 comprises path data 1108. The path data 1108 represents a path associated with the given route. The path data 1108 may represent the path with reference to one or more digital path markers. In this example, the route data 1112 comprises contextual data 1116. The contextual data 1116 represents contextual information associated with the given route. The contextual data 1116 may represent the contextual information with reference to one or more digital contextual markers. In this example, the route data 1112 comprises directional data 1118. The directional data 1118 represents directional information associated with the given route. The directional data 1118 may represent the directional information with reference to one or more digital directional markers. The route data 1112 data may comprise different data items in other examples. In other examples, the route data 1112 may be structured differently. For example, the contextual data 1116 and / or the directional data 1118 may be comprised in the path data 1108 in other examples. Digital Route Marker Definition As explained above, a digital route marker may represent one or more positions in 3D space. A position in 3D space may be represented by coordinates, such as longitude, latitude, and altitude. A digital route marker may represent a single point in 3D space. For example, such a digital route marker may represent specific longitude, latitude, and altitude coordinates. A digital route marker may represent a line. For example, such a digital route marker may represent specific longitude and latitude components and may not represent an altitude coordinate. As such, different digital route markers may be represented in different manners. In some examples, a digital route marker may be represented by a given set of coordinates when the digital route marker is defined, and a different set of coordinates may be used when the digital route marker is used (for example when a path-following person follows a path associated with the digital route marker). For example, a digital path marker may be defined with longitude, latitude, and altitude coordinates. When the digital path marker is used, the altitude coordinate may be disregarded such that the digital path marker is, in effect, a vertical line at the defined longitude and latitude coordinates. Alternatively, the digital path marker may be defined with longitude and latitude coordinates and may not be defined with an altitude coordinate. Recording more information than may be needed may consume additional storage space over what is strictly required, but provides additional flexibility for future usage. Usage can then be tailored, for example to guide path-following people in a highly ergonomic and intuitive manner. For a digital path marker that is used without an altitude component, a path-following person can be directed to the closest point on the vertical line to their user device. This enables path-following people of different heights to be guided using the same digital path marker while being able to hold their respective user devices horizontally. If, instead, digital path markers were used with an altitude component, path-following people of different heights may feel that they were being guided to the digital path markers at different inclines. For example, a particularly tall path-following person may feel that they are being guided downhill if they were guided based on a digital path marker that represents a waist-height position in 3D space for a particularly small path-following person. However, a digital contextual marker may be defined and used with longitude, latitude, and altitude coordinates. This enables the digital contextual marker to guide the path-following person to an altitude-dependent position in 3D space, such as a door handle. Without the altitude component, a path-following person would readily be able to locate the door handle in terms of longitude and latitude, but not necessarily in terms of altitude. A digital route marker may represent a region around one or more positions in 3D space. For example, a digital path marker may represent a vertical column instead of solely a vertical line. A path may be defined, in raw data, using one or more discrete digital path markers but may be represented, during guidance, as a continuous trail between the discrete digital path markers. A continuous trail may be easier and smoother to follow in some situations than a sequence of discrete digital path markers. However, discrete digital path markers may be ordered in a sequence, which can enable a path-following person to be guided in the correct direction. With the continuous trail example described above, a path-following person might inadvertently walk in the wrong direction along the trail, for example if they walk along a sub-path and return to a main path. RTK-GNSS, VO And VSLAM Real-time kinematic positioning satellite navigation, RTK-GNSS, uses surveying to correct errors in satellite navigation (GNSS) systems. A user device comprising an RTK-GNSS receiver may implement RTK-GNSS. RTK-GNSS enables high-precision localisation, for example with centimetre- level accuracy. Examples described herein may leverage RTK-GNSS to provide improved path- planning. RTK-GNSS enables accurate altitude coordinates to be recorded, in addition to longitude and latitude coordinates. Such altitude information may be used, for example, to record a highly accurate position in 3D space of a button for a road crossing, a door handle, even a particular elevator button on a panel of elevator buttons. Altitude may alternatively or additional be determined using computer vision. Visual odometry (VO) involves determining position and orientation by analysing camera data from one or more cameras. Visual SLAM (VSLAM or V-SLAM) involves using camera data from one or more cameras to perform SLAM, i.e. simultaneous localisation and map-generation in an unknown environment. Other Types Of Navigation Guidance Several existing applications guide a vehicle accurately. These include, but are not limited to, driverless cars, automated tractors, automated lawnmowers, and automated inspection and mapping drones (both flying and legged). Driverless cars typically exist in well-defined environments, namely roads. Such environments have GPS maps and, in some cases, high-resolution LIDAR data. Paths can be planned based on existing GPS data of known roads. Automated tractors and lawnmowers also only exist in specific settings, namely outdoors. They generally have user-set boundaries and paths. Automated vacuum cleaners similarly only exist in specific settings, namely indoors. However, because they are used indoors, they use different sensing techniques, such as VSLAM. Automated vacuum cleaners can overcome temporary obstacles. Legged and flying inspection drones fly along paths which are pre-set by humans. For a flying drone, the path may be a specific GPS path, or may be a trajectory as defined by VO in GPS-denied environments. Even in unfamiliar environments, flight-planning can be automated to avoid obstacles. However, drones that include RTK-GNSS and VO are too large to be handheld. In examples, one criterion for a path is trustworthiness. In addition, in examples, a path may go through an environment in which high-resolution, a priori mapping is unfeasible or otherwise unavailable. An example of such an environment is the home of a user. An example of a priori mapping is a LIDAR scan. In addition, GPS maps of streets exist. These cannot exist for certain environments, such as underground railway stations. Humans encounter an extraordinary number of edge cases that are not present in drone flight or autonomous road vehicles. Unlike on a road, there are no, or very few, defined street-walking rules. However, there are often many other nearby pedestrians. In addition, when walking, obstacles have many variations. For example, even in the case of a door, many variations exist. Examples of such variations include, but are not limited to, rotating doors, single doors, double doors, push doors, pull doors and automatic doors. In addition, most existing environments are unmapped. Some existing environments, such as open spaces, are unstructured. The above-indicated technology, namely RTK-GNSS, VO and VSLAM, has not been used in handheld user devices to guide a path-following person. In particular, navigation guidance devices are not available that can guide users with centimetre-accurate precision and over distances approaching those that may be travelled by an autonomous tractor or a driverless car. It might seem possible, in principle, to plan walking paths based on maps and GPS data. However, the large number of edge cases encountered by humans makes this unfeasible for centimetre-accurate guidance. Examples described herein provide a robust solution that does not rely on a pre-existing map, that can work with high precision (for example, 1 centimetre), over long distances (for example, 10 kilometres) and that can handle the large number of edge cases to which humans are subject. Guidance of this type and at this level does not currently exist. In particular, in some known haptic guidance examples, standard GPS is used to guide a person. However, this is at a high level, for example with greater than 1 metre accuracy, such as to guide a person to a street corner. In contrast, examples described herein restrict, or at least guide, a user to follow predefined paths. However, this magnifies robustness and makes it possible to use technologies such as RTK-GNSS to guide vision impaired people at a much high level of fidelity, for example lower than 10 centimetres, such to lift buttons and door handles. Examples described herein provide a handheld and / or wearable navigation guidance device that uses high-resolution localisation technology to enable long-range and high-precision human guidance. Such examples represent a significant departure from existing technologies that use GPS and VSLAM for short distances and / or low resolutions. Fallback Device-Localisation In some examples, a navigation guidance device as described herein may comprise a fallback device-localisation mechanism. The fallback device-localisation mechanism may be used in one or more predetermined fallback situations. An example of such a fallback device-localisation mechanism is VSLAM. An example of a predetermined fallback situation is when an RTK-GNSS signal is lost, such as when walking under a bridge. Alternatively, or additionally, a navigation guidance device may issue an alert if an RTK-GNSS signal is lost to indicate decreased localisation accuracy to a user. Initial Localisation In some initialisation situations, one type of high-precision localisation functionality may be unavailable, and a fallback initialisation mechanism may be used to localise a user device. For example, a user device may be initialised when a user is behind their front door at the start of a journey. An RTK-GNSS receiver may not be able to locate the user device accurately while the user device is indoors. In addition, it may take several minutes for the user device to be able to localise itself using an RTK-GNSS receiver once the user device is outdoors. This may limit effectiveness of the user device at the start of the journey. In some examples, the fallback initialisation mechanism is used to localise a user device during initialisation of the user device. For example, the fallback initialisation mechanism may be able to detect a physical marker in a known position on the floor and may be able to determine the position of the user device accordingly. In another example, the fallback initialisation mechanism may detect a fiducial marker to determine the position of the user device. In another example, the fallback initialisation mechanism may use VO to determine the position of the user device. In another example, a fallback initialisation mechanism may determine that the user device is docked in a docking station having a known position and may determine the position of the user device accordingly. The docking station may be an off-the-ground docking station. For example, the off-the- ground docking station may be a tabletop docking station, a wall-mounted docking station, or otherwise. In another example, the fallback initialisation mechanism may detect a radio-frequency identification (RFID) tag and may determine the position of the user device accordingly. This may provide a sufficiently accurate estimate for sensor fusion, for example. Temporary And / Or User-Specific Contextual Information A digital contextual marker and / or contextual information may be global (in other words, useable by all users), group-based (in other words, useable by a group of users), or user-specific (in other words, useable by one user). For example, a warning-related contextual information may be global such that all users can be warned of nearby danger. Contextual information may be group-based such that, for example, groups of users interested in a certain type of shop can be alerted to that type of shop being nearby. Contextual information may be user-based such that, for example, a given user may be reminded to purchase a particular grocery item when they visit a particular grocery shop. Contextual information may be temporary. For example, using the specific example of the user-specific grocery item reminder, the reminder could be set to expire at the end of a given day, after the user device has left the particular grocery shop, after the user has manually marked the reminder as completed, or otherwise. Although specific examples are described herein, other examples are envisaged. Such other examples may use one or more features from one or more examples described herein. Such other examples may use one or more features not described herein. Various numbered clauses of the present description follow. The claims of the present application as filed can be found after the heading “Claims”. i A sensory substitution device comprising: a spatial sensor configured to provide spatial sensor data indicative of a spatial representation of an environment in which the sensory substitution device is located; and a shape-change mechanism operable to cause the sensory substitution device to change shape, based on the spatial sensor data, to provide kinaesthetic output to a user of the sensory substitution device, wherein the kinaesthetic output is indicative of a target position in the environment. 2. A sensory substitution device according to clause 1, wherein the change of shape is in at least one degree of freedom. 3. A sensory substitution device according to clause 1 or 2, wherein the change of shape is in at least two degrees of freedom. 4. A sensory substitution device according to any of clauses 1 to 3, wherein the change of shape is in three degrees of freedom. 5. A sensory substitution device according to any of clauses 1 to 4, wherein the sensory substitution device comprises an elongate body and wherein the shape-change mechanism is operable to cause the elongate body of the sensory substitution device to change shape. 6. A sensory substitution device according to any of clauses 1 to 5, wherein at least a portion of the sensory substitution device is substantially cylindrical. 7. A sensory substitution device according to any of clauses 1 to 6, wherein the sensory substitution device comprises a gripping portion configured to be gripped by the user. 8. A sensory substitution device according to any of clauses 1 to 7, wherein the sensory substitution device is a handheld sensory substitution device. 9. A sensory substitution device according to any of clauses 1 to 8, wherein the sensory substitution device is configured to change shape based on closed-loop feedback, the closed-loop feedback based on a difference between the target position and a position in the environment at which the sensory substitution device is pointed. 10. A sensory substitution device according to any of clauses 1 to 9, wherein the change of shape comprises bending of the sensory substitution device. 11. A sensory substitution device according to any of clauses 1 to 10, wherein the shape-change mechanism is configured such that the sensory substitution device is substantially straight when the sensory substitution device is pointed at the target position. 12. A sensory substitution device according to any of clauses 1 to 11, wherein the change of shape comprises stretching and / or compressing of the sensory substitution device. 13. A sensory substitution device according to any of clauses 1 to 12, wherein the target position is a position determined using computer vision. 14. A sensory substitution device according to any of clauses 1 to 13, wherein the target position is a position determined using LIDAR. 15. A sensory substitution device according to any of clauses 1 to 14, wherein the target position is a position determined using ultrasound. 16. A sensory substitution device according to any of clauses 1 to 15, wherein the target position is a position determined using infrared. 17. A sensory substitution device according to any of clauses 1 to 16, wherein the target position is a position identified based on a voice input being processed using natural language processing. 18. A sensory substitution device according to any of clauses 1 to 17, wherein the sensory substitution device is operable to communicate verbally a distance from the sensory substitution device to the target position. 19. A sensory substitution device according to any of clauses 1 to 18, wherein the shape-change mechanism comprises elastomeric material. 20. A sensory substitution device according to any of clauses 1 to 19, wherein the shape-change mechanism is configured to cause the sensory substitution device to return to an equilibrium shape when the sensory substitution device is powered off. 21. A sensory substitution device according to any of clauses 1 to 20, wherein the sensory substitution device is configured to be controlled using SLAM. 22. A sensory substitution device according to any of clauses 1 to 21, wherein the device comprises object identification functionality. 23. A sensory substitution device according to any of clauses 1 to 22, wherein the sensory substitution device comprises a microcontroller and wherein the microcontroller is configured to control the shape-change mechanism based on the spatial sensor data. 24. A sensory substitution device according to any of clauses 1 to 23, wherein the sensory substitution device is operable to be controlled by at least one person other than the user while the user is using the sensory substitution device. 25, A sensory substitution device comprising: a sensor operable, in use, to output sensor data representative of an environment in which the sensory substitution device is being used by a user of the sensory substitution device; and a mechanism operable to cause the sensory substitution device to change shape, based on the output data, with motion that is not plane-constrained, whereby to provide proprioceptive output to the user. 26. A sensory substitution device comprising: a spatial sensor configured to provide spatial sensor data indicative of a spatial representation of an environment in which the sensory substitution device is located; and a shape-change mechanism operable to cause the sensory substitution device to change shape, based on the spatial sensor data, to provide kinaesthetic output to a user of the sensory substitution device, wherein the kinaesthetic output is indicative of a target position in the environment. 27. A sensory substitution device according to clause 26, wherein the change of shape is in at least one degree of freedom. 28. A sensory substitution device according to clause 26 or 27, wherein the change of shape is in at least two degrees of freedom. 29. A sensory substitution device according to any of clauses 26 to 28, wherein the change of shape is in three degrees of freedom. 30. A sensory substitution device according to any of clauses 26 to 29, wherein the sensory substitution device comprises an elongate body and wherein the shape-change mechanism is operable to cause the elongate body of the sensory substitution device to change shape. 31. A sensory substitution device according to any of clauses 26 to 30, wherein at least a portion of the sensory substitution device is substantially cylindrical. 32. A sensory substitution device according to any of clauses 26 to 31, wherein the sensory substitution device comprises a gripping portion configured to be gripped by the user. 33. A sensory substitution device according to any of clauses 26 to 32, wherein the sensory substitution device is a handheld sensory substitution device. 34. A sensory substitution device according to any of clauses 26 to 33, wherein the sensory substitution device is configured to change shape based on closed-loop feedback, the closed-loop feedback being based on a difference between the target position and a position in the environment at which the sensory substitution device is pointed. 35. A sensory substitution device according to any of clauses 26 to 34, wherein the change of shape comprises bending of the sensory substitution device. 36. A sensory substitution device according to any of clauses 26 to 35, wherein the shape-change mechanism is configured such that the sensory substitution device is substantially straight when the sensory substitution device is pointed at the target position. 37. A sensory substitution device according to any of clauses 26 to 36, wherein the change of shape comprises stretching and / or compressing of the sensory substitution device. 38. A sensory substitution device according to any of clauses 26 to 37, wherein the target position is a position determined using computer vision. 39. A sensory substitution device according to any of clauses 26 to 38, wherein the target position is a position determined using LIDAR. 40. A sensory substitution device according to any of clauses 26 to 39, wherein the target position is a position determined using ultrasound. 41. A sensory substitution device according to any of clauses 26 to 40, wherein the target position is a position determined using infrared. 42. A sensory substitution device according to any of clauses 26 to 41, wherein the target position is a position identified based on a voice input being processed using natural language processing. 43. A sensory substitution device according to any of clauses 26 to 42, wherein the sensory substitution device is operable to communicate verbally a distance from the sensory substitution device to the target position. 44, A sensory substitution device according to any of clauses 26 to 43, wherein the shape-change mechanism comprises elastomeric material. 45. A sensory substitution device according to any of clauses 26 to 44, wherein the shape-change mechanism is configured to cause the sensory substitution device to return to an equilibrium shape when the sensory substitution device is powered off. 46. A sensory substitution device according to any of clauses 26 to 45, wherein the sensory substitution device is configured to be controlled using SLAM. 47. A sensory substitution device according to any of clauses 26 to 46, wherein the sensory substitution device comprises object identification functionality. 48. A sensory substitution device according to any of clauses 26 to 47, wherein the sensory substitution device comprises a microcontroller and wherein the microcontroller is configured to control the shape-change mechanism based on the spatial sensor data. 49. A sensory substitution device according to any of clauses 26 to 48, wherein the sensory substitution device is operable to be controlled by at least one person other than the user while the user is using the sensory substitution device. 50. A sensory substitution device according to clauses 49, wherein the sensory substitution device is operable to be controlled by the at least one person based at least in part on video data transmitted to a device of the at least one person, the video data representing the environment in which the sensory substitution device is located. 51. A sensory substitution device according to any of clauses 26 to 50, wherein the sensory substitution device is operable to switch from tracking a first identified object to tracking a second identified object based on user input. 52. A sensory substitution device according to any of clauses 26 to 51, wherein the sensory substitution device is configured to perform tilt compensation to compensate for tilting of the sensory substitution device. 23. A sensory substitution device according to any of clauses 26 to 52, wherein the sensory substitution device is configured to provide the kinaesthetic output to the user of the sensory substitution device based at least in part on one or more digital markers in the environment. 54. A sensory substitution device according to any of clauses 26 to 53, wherein the sensory substitution device comprises a hand strap to hold a palm of the user against a body of the sensory substitution device. 55. A sensory substitution device according to any of clauses 26 to 54, wherein the sensory substitution device is configured to use one or more stimuli to indicate that the user is pointing the sensory substitution device within a given error bound of the target position. 56. A sensory substitution device according to any of clauses 26 to 54, wherein the sensory substitution device is configured to use one or more stimuli to indicate that the user is not pointing the sensory substitution device within a given error bound of the target position. 57. A sensory substitution device according to any of clauses 26 to 56, wherein the sensory substitution device comprises at least one tactile marker for at least one digit of at least one hand of the user. 58. A sensory substitution device according to any of clauses 26 to 57, wherein the kinaesthetic output is indicative of a path between the sensory substitution device and the target position, wherein the path corresponds to a biomechanically optimal path to the target position relative to a different, mathematically and / or spatially optimal path to the target position. 59. A sensory substitution device according to any of clauses 26 to 58, wherein the sensory substitution device comprises one or more light sources. 60. A sensory substitution device according to any of clauses 26 to 59, wherein the sensory substitution device is configured to alert the user in response to detecting a danger and / or a potential danger. 61. A sensory substitution device according to any of clauses 26 to 60, wherein the target position is a position determined using infrared. 62. A sensory substitution device according to any of clauses 26 to 61, wherein the sensory substitution device is configured to perform non-linear actuation, wherein an amount of shape change of the sensory substitution device correlates non-linearly to an amount of error between a pointing position and the target position. 63. A sensory substitution device comprising: a sensor operable, in use, to output sensor data representative of an environment in which the sensory substitution device is being used by a user of the sensory substitution device; and a mechanism operable to cause the sensory substitution device to change shape, based on the output data, with motion that is not plane-constrained, whereby to provide proprioceptive output to the user. 64. A sensory substitution device comprising: a shape-change mechanism operable to cause the sensory substitution device to change shape, based on spatial sensor data captured by a spatial sensor, to provide kinaesthetic output to a user of the sensory substitution device, wherein the kinaesthetic output is indicative of a target position. 65. A sensory substitution device comprising: a force-based mechanism operable to cause the sensory substitution device to provide kinaesthetic output to a user of the sensory substitution device based on spatial sensor data captured by a spatial sensor, wherein the kinaesthetic output is indicative of a target position. 66. A sensory substitution device comprising: a force-based mechanism operable to cause the sensory substitution device to provide kinaesthetic output to a user of the sensory substitution device, wherein the kinaesthetic output is indicative of a path between the sensory substitution device and a target position, and wherein the path corresponds to a biomechanically optimal path to the target position relative to a different, mathematically and / or spatially optimal path to the target position. 67. A sensory substitution device comprising: an elongate body; and amechanism operable to cause the sensory substitution device to provide kinaesthetic output to a user of the sensory substitution device based on spatial sensor data captured by a spatial sensor, wherein the kinaesthetic output directs the user to point the elongate body of the sensory substitution device at a target position. 68. A sensory substitution device comprising: amechanism operable to cause the sensory substitution device to provide kinaesthetic output to a user of the sensory substitution device; and a tilt compensator to compensate the kinaesthetic output for tiling of the sensory substitution device by the user. 69. A sensory substitution device comprising a mechanism operable to cause the sensory substitution device to provide kinaesthetic output to a user of the sensory substitution device, wherein the substitution device is operable to provide kinaesthetic output to guide a user of the sensory substitution device and is also operable to provide kinaesthetic output to alert the user to the presence of an object. 70. A shape-changing sensory substitution device configured to perform non-linear actuation, wherein an amount of shape change of the sensory substitution device correlates non-linearly to an amount of error between a pointing position and a target position. 71. A navigation guidance device comprising: input componentry configured to obtain path data representing a path along which a path- following person is to walk, the path data representing a digital path marker, the digital path marker being based on tracked movement of a path-defining person along a walked path, the walked path having been walked along by the path-defining person, wherein a digital path marker is a digital representation of a position in 3D space; and output componentry comprising a kinaesthetic output mechanism operable to provide kinaesthetic output to guide the path-following person along at least part of the path using the digital path marker. 22 A navigation guidance device according to clause 71, wherein the kinaesthetic output mechanism comprises a shape-change mechanism operable to cause the navigation guidance device to change shape to guide the path-following person along the at least part of the path. 73. A navigation guidance device according to clause 71 or 72, wherein the kinaesthetic output mechanism comprises a gyroscope. 74. A navigation guidance device according to any of clauses 71 to 73, wherein the path data represents a sequence of a discrete digital path markers. 75. A navigation guidance device according to clause 74, wherein the output componentry is configured to guide the path-following person along the at least part of the path by directing the path- following person to a position in 3D space represented by a next digital path marker in the sequence of digital path markers. 76. A navigation guidance device according to any of clauses 71 to 75, wherein the navigation guidance device comprises a handheld navigation guidance user device. 77. A navigation guidance device according to any of clauses 71 to 76, wherein the navigation guidance device comprises a wearable navigation guidance user device. 78. A navigation guidance according to any of clauses 71 to 77, wherein the navigation guidance device is configured to adjust the path in response to detecting a path adjustment trigger. 79. A navigation guidance device according to clause 78, wherein the navigation guidance device comprises a spatial sensor and wherein the path adjustment trigger comprises an obstacle along the path being detected based on spatial sensor data output by the spatial sensor. 80. A navigation guidance device according to clause 78 or 79, wherein the navigation guidance device is configured to generate adjusted path data in response to adjusting the path, the adjusted path data representing a digital path marker that is different from the digital path marker represented by the path data. 81. A navigation guidance device according to any of clauses 71 to 80, wherein the position in 3D space represented by the digital path marker is dependent on a location of the navigation guidance device relative to a body of the path-following person. 82. A navigation guidance device according to any of clauses 71 to 81, wherein the input componentry is configured to obtain contextual information data representing a digital contextual marker and wherein the output componentry is configured to guide the path-following person based on contextual information associated with the digital contextual marker. 83. A navigation guidance device according to clause 82, wherein the digital contextual marker is associated with a position in 3D space, wherein the digital contextual marker has an activation area, and wherein the navigation guidance device is configured to guide the path-following person based on the contextual information associated with the digital contextual marker when the navigation guidance device is located in the activation area. 84. A navigation guidance device according to any of clauses 71 to 83, wherein the input componentry is configured to obtain directional data representing a digital directional marker and wherein the output componentry is configured to guide the path-following person in a direction indicated by the digital directional marker. 85. A navigation guidance device according to any of clauses 71 to 84, wherein the output componentry is configured to illuminate some or all of the path. 86. A navigation guidance device according to any of clauses 71 to 85, wherein the digital path marker is a digital representation of multiple positions in 3D space, the multiple positions in 3D space having the same longitudinal and latitudinal components as each other and having different altitude components from each other. 87. A navigation guidance device according to clauses 86, wherein the output componentry is configured to direct the path-following person to a closest one of the multiple positions in 3D space when guiding the path-following person using the digital path marker. 88. A navigation guidance device according to any of clauses 71 to 87, wherein the navigation guidance device comprises a real-time kinematic positioning satellite navigation, RTK GNSS, receiver, the navigation guidance device being configured to use the RTK GNSS receiver to guide the path- following person. 89. A navigation guidance device according to clauses 88, wherein the navigation guidance device comprises a fallback initialisation mechanism, the fallback initialisation mechanism being useable to localise the navigation guidance device during initialisation of the navigation guidance device in response to the RTK GNSS receiver not being useable to localise the navigation guidance device. 90. A navigation guidance device according to any of clauses 71 to 89, wherein the navigation guidance device is configured to use visual simultaneous localization and mapping, VSLAM, to guide the path-following person. 91. A navigation guidance device according to any of clauses 71 to 90, wherein the navigation guidance device is configured to use visual odometry, VO, to guide the path-following person. 92. A navigation guidance device configured to obtain path data representing a path, the path data representing a digital path marker, the digital path marker being based on tracked movement of a path-defining person along a walked path, the walked path having been walked along by the path- defining person, wherein a digital path marker is a digital representation of a position in 3D space. 93. A navigation guidance device configured to use a real-time kinematic positioning satellite navigation, RTK-GNSS, receiver to provide navigation guidance to a user of the navigation guidance device, wherein the navigation guidance device comprises a handheld navigation guidance device and / or a wearable navigation guidance device. 94. A navigation guidance device configured to use visual simultaneous localization and mapping, VSLAM, to provide navigation guidance to a user of the navigation guidance device, wherein the navigation guidance device comprises a handheld navigation guidance device and / or a wearable navigation guidance device. 95. A navigation guidance device configured to use visual odometry, VO, to provide navigation guidance to a user of the navigation guidance device, wherein the navigation guidance device comprises a handheld navigation guidance device and / or a wearable navigation guidance device.

Claims

Claims 1: A navigation guidance device comprising: input componentry configured to obtain: path data representing a path along which a path-following person is to walk, the path data representing a digital path marker, the digital path marker being based on tracked movement of a path-defining person along a walked path, the walked path having been walked along by the path-defining person, wherein a digital path marker is a digital representation of a position in 3D space; and directional data representing a digital directional marker; and output componentry comprising a kinaesthetic output mechanism operable to provide kinaesthetic output to guide the path-following person along at least part of the path using the digital path marker, wherein the output componentry is configured to guide the path-following person in a direction indicated by the digital directional marker.

2. A navigation guidance device according to claim 1, wherein the kinaesthetic output mechanism comprises a shape-change mechanism operable to cause the navigation guidance device to change shape to guide the path-following person along the at least part of the path.

3. A navigation guidance device according to claim 1 or 2, wherein the kinaesthetic output mechanism comprises a gyroscope.

4. A navigation guidance device according to any of claims 1 to 3, wherein the path data represents a sequence of a discrete digital path markers.

5. A navigation guidance device according to claim 4, wherein the output componentry is configured to guide the path-following person along the at least part of the path by directing the path- following person to a position in 3D space represented by a next digital path marker in the sequence of digital path markers.

6. A navigation guidance device according to any of claims 1 to 5, wherein the navigation guidance device comprises a handheld navigation guidance user device.

7. A navigation guidance device according to any of claims 1 to 6, wherein the navigation guidance device comprises a wearable navigation guidance user device.

8. A navigation guidance according to any of claims 1 to 7, wherein the navigation guidance device is configured to adjust the path in response to detecting a path adjustment trigger.

9. A navigation guidance device according to claim 8, wherein the navigation guidance device comprises a spatial sensor and wherein the path adjustment trigger comprises an obstacle along the path being detected based on spatial sensor data output by the spatial sensor.

10. A navigation guidance device according to claim 8 or 9, wherein the navigation guidance device is configured to generate adjusted path data in response to adjusting the path, the adjusted path data representing a digital path marker that is different from the digital path marker represented by the path data.

11. A navigation guidance device according to any of claims 1 to 10, wherein the position in 3D space represented by the digital path marker is dependent on a location of the navigation guidance device relative to a body of the path-following person.

12. A navigation guidance device according to any of claims 1 to 11, wherein the input componentry is configured to obtain contextual information data representing a digital contextual marker and wherein the output componentry is configured to guide the path-following person based on contextual information associated with the digital contextual marker.

13. A navigation guidance device according to claim 12, wherein the digital contextual marker is associated with a position in 3D space, wherein the digital contextual marker has an activation area, and wherein the navigation guidance device is configured to guide the path-following person based on the contextual information associated with the digital contextual marker when the navigation guidance device is located in the activation area.

14. A navigation guidance device according to any of claims 1 to 13, wherein the output componentry is configured to illuminate some or all of the path.

15. A navigation guidance device according to any of claims 1 to 14, wherein the digital path marker is a digital representation of multiple positions in 3D space, the multiple positions in 3D space having the same longitudinal and latitudinal components as each other and having different altitude components from each other.

16. A navigation guidance device according to claim 15, wherein the output componentry is configured to direct the path-following person to a closest one of the multiple positions in 3D space when guiding the path-following person using the digital path marker.

17. A navigation guidance device according to any of claims 1 to 16, wherein the navigation guidance device comprises a real-time kinematic positioning satellite navigation, RTK GNSS, receiver, the navigation guidance device being configured to use the RTK GNSS receiver to guide the path- following person.

18. A navigation guidance device according to claim 17, wherein the navigation guidance device comprises a fallback initialisation mechanism, the fallback initialisation mechanism being useable to localise the navigation guidance device during initialisation of the navigation guidance device in response to the RTK GNSS receiver not being useable to localise the navigation guidance device.

19. A navigation guidance device according to any of claims 1 to 18, wherein the navigation guidance device is configured to use visual simultaneous localization and mapping, VSLAM, to guide the path-following person.

20. A navigation guidance device according to any of claims 1 to 19, wherein the navigation guidance device is configured to use visual odometry, VO, to guide the path-following person.

21. A navigation guidance device configured to obtain path data representing a path, the path data representing a digital path marker, the digital path marker being based on tracked movement of a path-defining person along a walked path, the walked path having been walked along by the path- defining person, wherein a digital path marker is a digital representation of a position in 3D space, wherein the navigation guidance device is configured to obtain directional data representing a digital directional marker, and wherein the navigation guidance device is configured to guide the path- following person in a direction indicated by the digital directional marker. Claims 1: A navigation guidance device comprising: input componentry configured to obtain path data representing a path along which a path- following person is to walk, the path data representing a digital path marker, the digital path marker being based on tracked movement of a path-defining person along a walked path, the walked path having been walked along by the path-defining person, wherein a digital path marker is a digital representation of a position in 3D space; and output componentry comprising a kinaesthetic output mechanism operable to provide kinaesthetic output to guide the path-following person along at least part of the path using the digital path marker, wherein the navigation guidance device comprises a real-time kinematic positioning satellite navigation, RTK GNSS, receiver, the navigation guidance device being configured to use the RTK GNSS receiver to guide the path-following person, and wherein the navigation guidance device comprises a fallback device-localisation mechanism, the fallback device-localisation mechanism being useable to localise the navigation guidance device in response to the RTK GNSS receiver not being useable to localise the navigation guidance device.

2. A navigation guidance device according to claim 1, wherein the kinaesthetic output mechanism comprises a shape-change mechanism operable to cause the navigation guidance device to change shape to guide the path-following person along the at least part of the path.

3. A navigation guidance device according to claim 1 or 2, wherein the kinaesthetic output mechanism comprises a gyroscope.

4. A navigation guidance device according to any of claims 1 to 3, wherein the path data represents a sequence of a discrete digital path markers.

5. A navigation guidance device according to claim 4, wherein the output componentry is configured to guide the path-following person along the at least part of the path by directing the path- following person to a position in 3D space represented by a next digital path marker in the sequence of digital path markers.

6. A navigation guidance device according to any of claims 1 to 5, wherein the navigation guidance device comprises a handheld navigation guidance user device.

7. A navigation guidance device according to any of claims 1 to 6, wherein the navigation guidance device comprises a wearable navigation guidance user device.

8. A navigation guidance according to any of claims 1 to 7, wherein the navigation guidance device is configured to adjust the path in response to detecting a path adjustment trigger.

9. A navigation guidance device according to claim 8, wherein the navigation guidance device comprises a spatial sensor and wherein the path adjustment trigger comprises an obstacle along the path being detected based on spatial sensor data output by the spatial sensor.

10. A navigation guidance device according to claim 8 or 9, wherein the navigation guidance device is configured to generate adjusted path data in response to adjusting the path, the adjusted path data representing a digital path marker that is different from the digital path marker represented by the path data.

11. A navigation guidance device according to any of claims 1 to 10, wherein the position in 3D space represented by the digital path marker is dependent on a location of the navigation guidance device relative to a body of the path-following person.

12. A navigation guidance device according to any of claims 1 to 11, wherein the input componentry is configured to obtain contextual information data representing a digital contextual marker and wherein the output componentry is configured to guide the path-following person based on contextual information associated with the digital contextual marker.

13. A navigation guidance device according to claim 12, wherein the digital contextual marker is associated with a position in 3D space, wherein the digital contextual marker has an activation area, and wherein the navigation guidance device is configured to guide the path-following person based on the contextual information associated with the digital contextual marker when the navigation guidance device is located in the activation area.

14. A navigation guidance device according to any of claims 1 to 13, wherein the output componentry is configured to illuminate some or all of the path.

15. A navigation guidance device according to any of claims 1 to 14, wherein the digital path marker is a digital representation of multiple positions in 3D space, the multiple positions in 3D space having the same longitudinal and latitudinal components as each other and having different altitude components from each other.

16. A navigation guidance device according to claim 15, wherein the output componentry is configured to direct the path-following person to a closest one of the multiple positions in 3D space when guiding the path-following person using the digital path marker.

17. A navigation guidance device according to any of claims 1 to 16, wherein the navigation guidance device is configured to use visual simultaneous localization and mapping, VSLAM, to guide the path-following person.

18. A navigation guidance device according to any of claims 1 to 17, wherein the navigation guidance device is configured to use visual odometry, VO, to guide the path-following person.

19. A navigation guidance device according to any of claims 1 to 18, wherein the input componentry is configured to obtain directional data representing a digital directional marker and wherein the output componentry is configured to guide the path-following person in a direction indicated by the digital directional marker.

Images