Semi-supervised intent recognition system and method

Abstract

An exoskeleton system is provided. A computer-implemented method for semi-supervised intent recognition for an exoskeleton system includes, in response to a state transition intent input, changing the exoskeleton system from operating in a first mode with sensitivity to detect state transitions at a first sensitivity level to operating in a second mode with sensitivity to detect state transitions at a second sensitivity level that is more sensitive than the first sensitivity level, identifying the state transition while operating in the second mode and using the second sensitivity level, and facilitating the identified state transition by actuating the exoskeleton system.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of U.S. Provisional Patent Application No. 62 / 551,696, filed August 29, 2017, which application is hereby incorporated by reference in its entirety and for all purposes.

[0002] This application is also related to U.S. Patent Application No. 15 / 953,296, filed April 13, 2018, related to U.S. Patent Application No. 15 / 887,866, filed February 2, 2018, related to U.S. Patent Application No. 15 / 823,523, filed November 27, 2017, and related to U.S. Patent Application No. 15 / 082,824, filed March 28, 2016, which applications are also incorporated herein by reference in their entirety and for all purposes. [Brief explanation of the drawings]

[0003] [Figure 1] 1 is an illustration of an example embodiment of an exoskeleton system worn by a user. [Figure 2] 1 is an illustration of an example of another embodiment of an exoskeleton system worn by a user while skiing. [Figure 3] 10 is an illustration of an example of a further embodiment of an exoskeleton system worn by a user while skiing. [Figure 4] 10 is an illustration of an example of yet a further embodiment of an exoskeleton system worn on the leg of a user. [Figure 5] FIG. 1 is a block diagram illustrating an embodiment of an exoskeleton system. [Figure 6] 1 illustrates an example state machine for an exoskeleton system including multiple system states and transitions between system states. [Figure 7] 7 is an example of a fully supervised intent recognition method illustrated in the context of the state machine of FIG. 6 and a user interface with a first button. [Figure 8] 7 is another example of a fully supervised intent recognition method illustrated in the context of a user interface having the state machine and first and second buttons of FIG. 6; [Figure 9] 7 is a further example of a fully supervised intent recognition method illustrated in the context of the state machine of FIG. 6 and a user interface with a first button. [Figure 10] 1 illustrates an example of an unsupervised intent recognition method according to one embodiment. [Figure 11] 1 illustrates an example embodiment of a semi-supervised intent recognition method. [Figure 12] 1 illustrates an example state machine for a supervised intent recognition method in which the standing state has eight possible transitions to eight respective states, and buttons are mapped to a single transition and state pair. [Figure 13] 1 illustrates an example state machine for supervised intent recognition where the standing state has eight possible transitions to eight respective states, and four buttons are each mapped to a single transition and state pair. [Figure 14] 14 illustrates an embodiment of the semi-supervised intent recognition method with the state machine shown in FIGS. 12 and 13 and a user interface with a single button for indicating intent to make a state transition. [Figure 15] FIG. 1 is a block diagram of a semi-supervised intent recognition method according to one embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0004] It should be noted that the drawings are not drawn to scale, and that elements of similar structure or function are generally represented by the same reference numerals throughout the drawings for illustrative purposes. It should also be noted that the drawings are intended only to facilitate the description of preferred embodiments. The drawings do not illustrate every aspect of the described embodiments and do not limit the scope of the present disclosure.

[0005] This application discloses example embodiments relating to the design and implementation of novel systems and methods for recognition of a user's intent in an enhanced exoskeleton. In various embodiments, methods for intent recognition in exoskeleton-type devices in their simplest form can allow users to provide direct indication of their intent through manual input (e.g., by using buttons), while other methods can be designed to remove reliance on direct interaction for standard operation. This disclosure, in various examples, describes systems and methods that allow users to provide input that can improve the accuracy of the device's intent recognition without making the device subordinate to user commands.

[0006] The embodiments described herein provide substantial improvements over alternative methods of intent recognition in exoskeletons. For example, one alternative method for intent recognition in exoskeletal devices is a fully supervised approach that provides the user or device proctor (e.g., a physical therapist) with the ability to directly indicate desired changes in the user's intended behavior. These methods can directly tie triggers in the device's state behavior to manual input, but this requires the user to indicate a wide variety of behaviors, such as sitting, standing, climbing or descending stairs, walking, and running.

[0007] In an effort to reduce the user's direct involvement in the device's behavior, alternative unsupervised methods have been developed that use device sensors to automatically determine the user's intended maneuvers without direct interaction from the operator. While this offers the potential to reduce the burden on the user and increase the number of operational modes the device can recognize, it introduces the risk of erroneous intent recognition. While several methods have been developed that use a combination of automated and direct identification, they still present the same challenges. The various systems and methods described herein allow the operator to provide information directly to the device without the operator having to directly manipulate the exoskeleton's operational state.

[0008] This disclosure teaches how to develop various embodiments of semi-supervised intent recognition methods. This approach can allow an operator to provide a device with additional information that the device can use to manipulate its behavior. In these embodiments, a user can provide direct input to a machine to improve its decision-making capabilities without directly dictating the decisions to be made.

[0009] In some embodiments, the direct input provided by the user is not directly correlated with a specific maneuver (e.g., taking an action). Another embodiment provides the user with only a single input of a direct indication of intent through the origin of a button. In this embodiment, the button does not correlate with a specific maneuver, such as walking, running, or standing. Instead, the button merely indicates the user's desire or intent to change behavior. In one example, if the operator is walking and intends to transition to standing or running, the user need only push the state transition intent button to alert the device to anticipate the potential decision. In such an embodiment, the device's behavior is not fully supervised or directly driven by the user's indicated intent. Instead, when the user specifies that a change in intent is likely to occur, the device can implement a method that is more responsive to user behavior and then responds accordingly, such as assisting the user in physically transitioning from walking to running or standing, or doing nothing.

[0010] In a further embodiment, direct input provided by a user can be directly correlated with a particular maneuver, but the device is still not directly operated by an indicated intent. An embodiment may include a user wearing a device in a seated position, with the device having a single button. In this position, even if a button is typically used to represent a change in indicated behavior, the only valid change in behavior for this device in this embodiment from a seated position is standing up. As a result, a single button press can be directly correlated with a single maneuver, such as a transition from sitting to standing. However, in this embodiment, the device may not be directly supervised by the user's indicated intent. As a result, the device does not change behavior immediately upon or in direct response to the button press, but instead becomes more sensitive to detecting a transition from sitting to standing initiated by the user.

[0011] The exoskeleton system can respond to the operator's semi-supervised intent in various ways. In one embodiment, the exoskeleton device can use an indication of intent from the operator to begin monitoring device sensors to look for changes in behavior. In another embodiment, the indicated intent can be used to increase the sensitivity of an already operational set of unsupervised intent recognition methods. This can be done by allowing the exoskeleton device to lower the reliance required to initiate a change in intended maneuver. In yet another embodiment, the indicated intent can be treated as just another sensor input. The device can then provide the user's indicated intent according to device sensors to a traditional unsupervised intent recognition method. This may be desirable in cases using data-driven intent recognition algorithms that utilize machine learning algorithms to infer appropriate points of change in intent. It is important to note that the previously described embodiments are illustrative but do not encompass all potential additional embodiments that can utilize semi-supervised indications of user intent and therefore should not be construed as limiting.

[0012] In various embodiments, a user can provide semi-supervised manual indications of intent to the exoskeleton device through various input methods. The origin of the input method does not limit or constrain the application of the disclosed systems and methods when incorporating semi-supervised input to form a better estimate of the user's intended behavior. Some potential input methods include, but are not limited to, physical buttons on the device, unique button presses (e.g., double-click or long press), individual gestures (e.g., arm wave, foot stomp), spoken commands, a mobile device interface, and interpreted manual input through another sensor input (e.g., inferring to knock on the device through observing IMU signals), etc.

[0013] For purposes of clarity, example embodiments are discussed in the context of the design and implementation of a skeletal system (e.g., as shown in FIG. 1 ), but the systems and methods described and illustrated herein can have application to a wide variety of worn devices in which the device uses on-board sensors for the purpose of recognizing a user's intended actions. A particular example of this is footwear, and in particular active footwear, where the device uses included sensors to determine the operator's intended actions so that it can report statistics to the user and adapt performance characteristics. In these applications, designers are challenged by the same issues surrounding balancing the stability of fully supervised intent recognizers with the availability of unsupervised options. The application of semi-supervised methods as disclosed herein can be a solution to balancing these needs in other augmented wearable devices as well.

[0014] 1, an example embodiment of an exoskeleton system 100 is illustrated as worn by a human user 101. As shown in this example, the exoskeleton system 100 comprises left and right leg actuator units 110L, 110R coupled to the user's left and right legs 102L, 102R, respectively. While in this example illustration, a portion of the right leg actuator unit 110R is obscured by the right leg 102R, it should be clear that in various embodiments, the left and right leg actuator units 110L, 110R can be substantially mirror images of one another.

[0015] The leg actuator unit 110 can include an upper arm 115 and a forearm 120 rotatably coupled via a joint 125. A bellows actuator 130 extends between plates 140 coupled at each end of the upper arm 115 and the forearm 120, with the plates 140 coupled to separate rotatable portions of the joint 125. A plurality of constraining ribs 135 extend from the joint 125 and surround a portion of the bellows actuator 130, as described in further detail herein. One or more sets of air tubing 145 can be coupled to the bellows actuator 130 to introduce and / or remove fluid from the bellows actuator 130, causing it to expand and contract, as discussed herein.

[0016] The leg actuator units 110L, 110R can be coupled around the legs 102L, 102R of the user 101 via one or more couplers 150 (e.g., straps that encircle the legs 104) to joints 125 positioned at the knees 103L, 103R of the user 101, with the upper arms 115 of the leg actuator units 110L, 110R coupled around the thighs 104L, 104R of the user 101. The forearms 120 of the leg actuator units 110L, 110R can be coupled around the lower legs 105L, 105R of the user 101 via one or more couplers 150. 1, the upper arm 115 can be coupled to the thigh 104 of the leg 102 above the knee 103 via two couplers 150, and the forearm 120 can be coupled to the lower leg 105 of the leg 102 below the knee 103 via two couplers 150. It is important to note that some of these components can be omitted in certain embodiments, some of which are discussed herein. Additionally, in further embodiments, one or more of the components discussed herein can be operatively replaced with alternative structures to produce the same functionality.

[0017] As discussed herein, the exoskeleton system 100 can be configured for a variety of suitable uses. For example, FIGS. 2 and 3 illustrate an exoskeleton system 100 used by a user while skiing. As shown in FIGS. 2 and 3, the user can wear the exoskeleton system 100 and a ski assembly 200 including a pair of ski boots 210 and a pair of skis 220. In various embodiments, the forearm 120 of the leg actuator unit 110 can be removably coupled to the ski boot 210 via a coupler 150. Such an embodiment may be desirable for directing force from the leg actuator unit 110 to the ski assembly. For example, as shown in FIGS. 2 and 3, the coupler 150 at the distal end of the forearm 120 can couple the leg actuator unit 110 to the ski boot 210, and the coupler 150 at the distal end of the upper arm 115 can couple the leg actuator unit 110 to the thigh 104 of the user 101.

[0018] The upper arms and forearms 115, 120 of the leg actuator units 110 can be coupled to the legs 102 of the user 101 in a variety of suitable manners. For example, FIG. 1 illustrates an embodiment in which the upper arms and forearms 115, 120 of the leg actuator units 110 and joints 125 are coupled along the sides of the top and bottom 104, 105 of the legs 102. FIGS. 4a and 4b illustrate another embodiment of the exoskeleton system 100 in which the joints 125 are positioned laterally adjacent the knee 103, with the axis of rotation K of the joints 125 aligned with the axis of rotation of the knee 103. The upper arms 115 can extend from the joints 125 to the front of the thighs 104 along the sides of the thighs 104. A portion of the upper arms 115 on the front of the thighs 104 can extend along axis U. The forearm 120 can extend along the side of the lower leg 105 from a medial position at the joint 125 to a posterior position at the bottom end of the lower leg 105, with a portion extending along an axis L perpendicular to the axis K.

[0019] In various embodiments, the joint structure 125 can constrain the bellows actuator 130 so that the force generated by the actuator fluid pressure within the bellows actuator 130 can be directed around an instantaneous center (which may or may not be fixed in space). In some cases of an outward-winding or rotary joint 125, or a body sliding on a curved surface, this instantaneous center can coincide with the instantaneous center of rotation of the joint 125 or the curved surface. The force generated by the leg actuator unit 110 around the rotary joint 125 can be used to apply a moment about the instantaneous center and can also be used to apply a directed force. In some cases of prismatic or linear joints (e.g., a slide on a rail), the instantaneous center can be kinematically considered to be located at infinity, in which case forces directed around this infinite instantaneous center can be considered as forces directed along the axis of motion of the prismatic joint. In various embodiments, it can be sufficient for the rotary joint 125 to be constructed from a mechanical pivot mechanism. In such embodiments, joint 125 can have a fixed center of rotation that can be easy to define, and bellows actuator 130 can move relative to joint 125. In further embodiments, it can be advantageous for joint 125 to include a compound linkage that does not have a single fixed center of rotation. In yet another embodiment, joint 125 can comprise a curved design that does not have a fixed joint pivot. In still further embodiments, joint 125 can comprise a structure such as a human joint or a robotic joint.

[0020] In various embodiments, the leg actuator unit 110 (e.g., comprising the bellows actuator 130, the joint structure 125, and the constraining rib 135, etc.) can be integrated into a system to use the generated directed force of the leg actuator unit 110 to accomplish various tasks. In some examples, the leg actuator unit 110 can have one or more advantageous advantages when the leg actuator unit 110 is configured to assist the human body or is included in the augmented-exoskeleton system 100. In example embodiments, the leg actuator unit 110 can be configured to assist the movement of a human user about the user's knee joint 103. To do so, in some examples, the instant center of the leg actuator unit 110 can be designed to coincide or nearly coincide with the instant center of rotation of the knee (e.g., aligned along a common axis K, as shown in FIG. 4a). In one example configuration, the leg actuator unit 110 can be positioned laterally (as opposed to anteriorly or posteriorly) relative to the knee joint 103, as shown in FIGS. 1, 2, 3, and 4a. In alternative embodiment configurations, the leg actuator unit 110 can be positioned behind the knee 103, in front of the knee 103, inside the knee 103, etc. In various embodiments, the human knee joint 103 can function as (in addition to or instead of) joint 125 of the leg actuator unit 110.

[0021] For clarity, the example embodiments discussed herein should not be viewed as limitations on the potential applications of the leg actuator unit 110 described within this disclosure. The leg actuator unit 110 can be used on other joints of the body, including, but not limited to, the elbow, hip, fingers, spine, or neck, and in some embodiments, the leg actuator unit 110 can be used in applications other than on the human body, such as in a robot for general-purpose actuation.

[0022] Some embodiments can adapt the configuration of the leg actuator unit 110 as described herein for linear actuation applications. In an example embodiment, the bellows 130 can comprise a two-layer impermeable / non-expandable structure, and one end of the constraining rib 135 can be fixed to the bellows 130 at a predetermined position. The joint structure 125 in various embodiments can be configured as a series of slides on a pair of linear guide rails, with the remaining end of each constraining rib 135 connected to the slide. Thus, the movement and force of the fluid actuator can be constrained and directed along the linear rails.

[0023] FIG. 5 is a block diagram of an example embodiment of an exoskeleton system 100 including an exoskeleton device 510 operably connected to a pneumatic system 520. The exoskeleton device 510 includes a processor 511, a memory 512, one or more sensors 513, a communication unit 514, and a user interface 515. A plurality of actuators 130 are operably coupled to the pneumatic system 520 via respective pneumatic lines 145. The plurality of actuators 130 include a pair of knee actuators 130L, 130R positioned on the right and left sides of the body 100. For example, as discussed above, the example exoskeleton system 100 shown in FIG. 5 can include left and right leg actuator units 110L, 110R on each side of the body 101 as shown in FIGS. 1-3.

[0024] In various embodiments, example system 100 can be configured to move and / or enhance the movement of a user wearing exoskeleton system 110. For example, exoskeleton device 510 can provide instructions to pneumatic system 520, which can selectively inflate and / or deflate bellows actuator 130 via air tubing 145. Such selective inflation and / or deflation of bellows actuator 130 can produce body movements such as walking, running, jumping, climbing, lifting, throwing, squatting, or skiing, and / or augment movement of one or both legs 102. In further embodiments, pneumatic system 520 can be manually controlled, configured to apply a constant pressure, or operate in any other suitable manner.

[0025] In some embodiments, such movements can be controlled and / or programmed by the user 101 wearing the exoskeleton system 100 or by another human. In some embodiments, the exoskeleton system 100 can be controlled by the user's movements. For example, the exoskeleton device 510 can sense that the user is walking and carrying a load and can provide augmented assistance to the user via the actuators 130 to reduce the load and the work associated with walking. Similarly, if the user 101 wears the exoskeleton system 100 while skiing, the exoskeleton system 100 can sense the movements of the user 101 (e.g., made by the user 101 or in response to the terrain, etc.) and can provide augmented assistance to the user via the actuators 130 to increase or provide assistance to the user while skiing.

[0026] Thus, in various embodiments, the exoskeleton system 130 can react automatically without direct user interaction. In further embodiments, movements can be controlled in real time by a controller, joystick, or through a control. Additionally, some movements can be pre-programmed and selectively triggered instead of being fully controlled (e.g., walk forward, sit down, bend down). In some embodiments, movements can be controlled by generalized commands (e.g., walk from point A to point B, pick up a box from shelf A and move it to shelf B).

[0027] In various embodiments, the exoskeleton device 100 can be operable to perform methods or portions of methods described in further detail below or in the related applications incorporated herein by reference. For example, the memory 512 can include non-permanent computer-readable instructions that, when executed by the processor 511, cause the exoskeleton system 100 to perform methods or portions of methods described herein or in the related applications incorporated herein by reference. The communication unit 514 can include hardware and / or software that enables the exoskeleton system 100 to communicate, directly or over a network, with other devices, including user devices, classification servers, or other exoskeleton systems.

[0028] In some embodiments, the sensors 513 may include any suitable type of sensors, and the sensors 513 may be located in a central location or may be distributed around the exoskeleton system 100. For example, in some embodiments, the exoskeleton system 100 may include multiple accelerometers, force sensors, position sensors, pressure sensors, etc., at various suitable locations, including the arms 115, 120, the joints 125, the actuators 130, or any other location. Thus, in some examples, the sensor data may correspond to the physical state of one or more actuators 130, the physical state of portions of the exoskeleton system 100, the physical state of the exoskeleton system 100 as a whole, etc. In some embodiments, the exoskeleton system 100 may include a global positioning system (GPS), a camera, a range sensing system, or environmental sensors.

[0029] The user interface 515 may include various suitable types of user interfaces, including one or more of physical buttons, a touch screen, a smartphone, a tablet computer, a wearable device, etc. For example, in some embodiments, the exoskeleton system 100 may comprise an embedded system including the user interface 515 or the exoskeleton device 510 that may be operably connected to a separate device (e.g., a smartphone) via a wired or wireless communication network (e.g., Bluetooth, Wi-Fi, the Internet, etc.).

[0030] Pneumatic system 520 may comprise any suitable device or system operable to individually or collectively inflate and / or deflate actuators 130. For example, in one embodiment, pneumatic system 520 may comprise a diaphragm compressor such as disclosed in related patent application Ser. No. 14 / 577,817, filed Dec. 19, 2014, and / or a poppet valve system such as described in related U.S. patent application Ser. No. 15 / 083,015, filed Mar. 28, 2016, which issued as U.S. Patent No. 9,995,321.

[0031] As discussed herein, various suitable exoskeletal systems 100 can be used in various suitable ways and for various suitable applications. However, such examples should not be construed as limiting the wide variety of exoskeletal systems 100, or portions thereof, that are within the scope and spirit of the present disclosure. Accordingly, exoskeletal systems 100 that are more or less complex than the examples of Figures 1, 2, 3, 4a, 4b, and 5 are within the scope of the present disclosure.

[0032] Additionally, while various embodiments relate to exoskeleton system 100 associated with a user's legs or lower body, further embodiments may relate to any suitable portion of a user's body, including the torso, arms, head, or legs. It should also be clear that while various embodiments relate to exoskeletons, the present disclosure may be applied to other similar types of technology, including prosthetics, body implants, or robots. Furthermore, some embodiments may relate to a human user, while other embodiments may relate to an animal user, a robotic user, or various types of mechanical devices.

[0033] As discussed herein, various embodiments relate to methods of semi-supervised intent recognition for wearable devices such as exoskeleton system 100. The semi-supervised intent recognition methods of various embodiments can be distinguished from fully supervised and unsupervised intent recognition methods, as described in further detail below.

[0034] Turning to Figure 6, an example state machine 600 for the exoskeleton system 100 is illustrated, including multiple system states and transitions between system states. More specifically, the state machine 600 is shown with a sitting state 605, from which the exoskeleton system 100 can transition to a standing state 615 via a sitting-to-standing transition 610. The exoskeleton system 100 can transition from the standing state 615 to a standing state 625 via a standing-to-standing transition 620. The exoskeleton system 100 can transition from the standing state 625 to a sitting state 635 via a standing-to-sitting transition 630. The exoskeleton system 100 can transition from the sitting state 635 to the sitting state 605 via a sitting-to-sitting transition 640.

[0035] For example, when the user 101 is sitting in a chair, the exoskeleton system 100 is in a sitting state 605, and when the user 101 wishes to stand up, the exoskeleton system 100 can move from sitting 605 to standing 620 via a standing state 615, which moves the user 101 from a sitting position to a standing position. When the user 101 is standing by the chair, the exoskeleton system 100 is in a standing state 625, and when the user 101 wishes to sit in the chair, the exoskeleton system 100 can move from standing 625 to sitting 605 via a sitting state 635, which moves the user 101 from a standing position to a sitting position.

[0036] Also, as shown in state machine 600, exoskeletal system 100 can move from standing state 625 to walking state 650 via standing to walking transition 645. Exoskeletal system 100 can move from walking state 650 to standing state 625 via walking to standing transition 655. For example, if user 101 is standing 625, user 101 can select walking 650 and can select to stop walking 650 and return to standing 625.

[0037] The example state machine 600 is used herein for illustrative purposes only and should not be construed as limiting the wide variety of state machines for the exoskeleton system 200 that are within the scope and spirit of the present disclosure. For example, some embodiments may include a simpler state machine having only standing and walking states 625, 650. Further embodiments may include additional states, such as a running state from the walking state 650.

[0038] Turning to FIG. 7, an embodiment of a fully supervised intent recognition method 700 is illustrated in the context of the state machine 600 of FIG. 6 and a user interface 515 (see FIG. 5) having an A button 710. In the fully supervised state machine of various embodiments, the user 101 provides direct manual input to the interface 515 commanding the initiation of a single unique transition from one state to another, and upon the transition, the exoskeleton system 100 is subordinated to initiate that transition. In this embodiment, that manual input is represented by a button press of the A button 710. The A button 710 is shown mapped to a single transition from the standing state 625 to the walking state 650 (i.e., the standing to walking transition 645). When button A is pressed and the exoskeleton system 100 detects that the user 101 is in a stable configuration to initiate the transition to walking 650, the exoskeleton system 100 initiates the transition 645 from the standing state 625 to the walking state 650. In other words, in this example, button A can only trigger the standing to walking transition 645 from standing state 625 to walking state 650, and all other transitions (i.e., 610, 620, 630, 640, 655) are not available via a button press of button A 710. This transition, if successfully completed, results in the device wearer physically transitioning from standing to walking in this example.

[0039] Turning to Figure 8, an embodiment of a fully supervised intent recognition method 800 is illustrated in the context of the state machine 600 of Figure 6 and a user interface 515 having first and second buttons 710, 720. More specifically, extending the embodiment of Figure 7 to accommodate multiple transitions in a fully supervised intent recognition system, button A is mapped to a single transition 645 from the standing state 625 to the walking state 650, as discussed above. In addition, button B 720 is shown mapped to a single transition from the sitting state 605 to the standing state 615 (i.e., the sitting to standing transition 610).

[0040] As discussed herein, if the A button 710 is pressed and the user 101 is stable, the exoskeleton system 100 initiates a transition from standing 625 to walking 650. If the B button 720 is pressed, the exoskeleton system 100 initiates a sitting-to-standing transition 610 from sitting 605 to standing 615, causing the user 101 to stand up from sitting. From there, the exoskeleton system can then interpret whether the user 101 has fully transitioned from the standing state 615 to the standing state 625 via the device-interpreted standing-to-standing transition 620; if not, the sitting-to-standing transition 610 can be aborted as a stability measurement and the user can return to sitting. In other words, pressing the B button 720 on the interface 515 can trigger a sitting to standing transition 610 from sitting 605 to standing state 615, and the exoskeleton device 100 then transitions 620 to standing state 625 unless an error occurs, in which case the device returns to sitting state 605.

[0041] Thus, the A button 710 can only trigger the standing to walking transition 645 from the standing state 625 to the walking state 650, and the B button 720 can only trigger the sitting to standing transition 610 from the sitting state 605 to the standing state 615; all other transitions (i.e., 620, 630, 640, 655) are not available via button presses of the A button 710 or the B button 720.

[0042] Turning to Figure 9, another embodiment of a fully supervised intent recognition method 900 is illustrated in the context of the state machine 600 of Figure 6 and the user interface 515 (see Figure 5) having an A button 710. In particular, Figure 9 illustrates another variation of the fully supervised state machine 900 in which the A button 710 is mapped such that if the exoskeleton system 100 is in a standing state 625 and the user 101 is stable, pressing the A button 710 causes the exoskeleton system 100 to initiate a standing-to-walking transition 645 to a walking state 650, and if the exoskeleton system 100 is in a sitting state 605 and the user 101 is stable, the exoskeleton system 100 initiates a sitting-to-standing transition 610 to a standing state 615, after which the exoskeleton system 100 then interprets whether a successful transition 620 to the standing state 625 was made and acts accordingly. The button configuration in this embodiment is similar to the previous embodiment of Figure 8 with both buttons A and B 710, 720, except that the same button 710 is mapped to two specific transitions 610, 645, respectively, instead of one transition. Thus, in this embodiment of the fully supervised intent recognition method, a single button press is mapped to one transition, and one transition only, regardless of whether one, two, or multiple buttons are used to indicate that button press.

[0043] Fully supervised intent recognition methods, such as those discussed above, can be distinguished from unsupervised intent recognition methods. For example, FIG. 10 illustrates an example of an unsupervised intent recognition method. More specifically, FIG. 10 illustrates an unsupervised state machine 1000 in which the user 101 does not provide direct manual input to the exoskeleton system 100's intent recognition. Instead, the exoskeleton system 100 continuously monitors sensor inputs and interprets which state the exoskeleton system 100 is currently in and which transition the user 101 is attempting to initiate. When a threshold for a possible transition from the currently detected state is reached based on sensor data (e.g., from sensors 513) and the user 101 is interpreted as being in a stable configuration, the exoskeleton system 100 can then initiate the interpreted transition.

[0044] In contrast to the fully supervised intent recognition method discussed in FIGS. 7-9 , each of the transitions 610, 620, 630, 640, 645, and 655 shown in FIG. 10 is a device-interpreted transition, and the exoskeleton system 100 determines the current state (i.e., sitting 605, standing 615, standing 625, sitting 635, and walking 650) and determines which transition, if any, the user is attempting to initiate. Accordingly, the user interface 515 of the example of FIG. 10 does not have buttons or other elements or mechanisms that allow the user 101 to initiate one or more particular transitions (although the user interface 515 can have other suitable functionality). In other words, the unsupervised method of FIG. 10 does not allow the user 101 to provide input indicating a desire to make or initiate a transition, whereas the supervised intent recognition method discussed in FIGS. 7-9 allows the user 101 to initiate transitions to some or all states through the user interface 515.

[0045] As discussed herein, fully supervised and unsupervised intent recognition methods can be distinguished from semi-supervised intent recognition methods, which are described in more detail below. For example, FIG. 11 illustrates an example embodiment of a semi-supervised intent recognition method. In particular, FIG. 11 illustrates a semi-supervised state machine 1100 in which a user 101 provides direct manual input to the exoskeleton system 100's intent recognition, indicating that the exoskeleton system 100 should explore state transitions from a current state, where the current state is known to or determined by the exoskeleton system 100 at the time of the manual input by the user 101.

[0046] Such increased observation of state transitions can be achieved in various suitable manners, such as by lowering one or more thresholds for interpreting whether a transition has occurred, which can increase the likelihood that the transition will be observed from the sensor input (e.g., from sensor data received from sensor 513).

[0047] If a state transition is detected after a manual input (e.g., in this example, pressing button X 1130), the exoskeleton system 100 then proceeds to initiate the detected state transition. However, if a state transition is not detected, the exoskeleton system 100 takes no action and after a predefined timeout, the exoskeleton system 100 stops searching for a transition and returns the exoskeleton system 100 to a normal state where it waits for the next manual input.

[0048] In other words, in various embodiments, the exoskeleton system 100 can monitor and respond to movements of the user 101 in standard operating conditions, including identifying and initiating various state transitions (e.g., any possible state transitions as shown in the example of FIG. 11 ) along with identifying state transitions associated with a first set of one or more thresholds or criteria, etc. In response to input from the user (e.g., pressing a single button X 1130), the exoskeleton system 100 can still monitor and respond to movements of the user 101, but according to a second set of one or more thresholds or criteria, etc., such that identifying state transitions is easier compared to standard operation under the first set.

[0049] More specifically, for some sets of sensor data, a given state transition is not identified as existing when the exoskeleton system 100 is operating under a first set, but is identified as existing under a second set of one or more thresholds or criteria, etc. Thus, in various embodiments, the user 101 can provide a given input (e.g., pressing a single button X 1130) to cause the exoskeleton system 100 to become more sensitive to identifying the state transition.

[0050] In various embodiments, sensitivity to state transitions initiated by the user 101 can be based on possible state transitions given the state the user 101 and the exoskeleton system 100 are currently in. Thus, in various embodiments, after an indication of an intent to make a state change is received (e.g., via the user 101 pressing the X button 1130), a determination can be made as to what state the user 101 and the exoskeleton system 100 are currently in, and sensitivity to potential state changes by the user 101 can be adjusted based on the determined current state.

[0051] 11 , if a determination is made that the user is in a sitting state 605, the sensitivity to identifying a transition to a standing state 615 can be adjusted to be more sensitive, and other states that are not directly reachable from the sitting state (e.g., a walking state 650 or a sitting state 635) can be eliminated as potential states that may be detected or identified. Additionally, if multiple state transitions are possible from a given state, the sensitivity can be adjusted for those multiple potential state transitions. For example, with reference to FIG. 11 , if a determination is made that the user is in a standing state 625, the sensitivity to identifying a transition to a sitting state or a walking state 635, 650 can be adjusted to be more sensitive, and other states that are not directly reachable from the sitting state (e.g., standing 615) can be eliminated as potential states that may be detected or identified.

[0052] Making the exoskeleton system 100 more sensitive to state transitions in response to input from the user 101 may be desirable to improve the experience of a user wearing the exoskeleton system 100. For example, during standard operation, the threshold for identifying and responding to state transitions may be higher to prevent false positives of state transitions, and may also allow the exoskeleton system 100 to respond when a state transition should occur.

[0053] However, if the user intends to initiate a state transition (e.g., moving from sitting to standing, moving from standing to sitting, or moving from standing to walking, etc.), the user 101 can provide an input indicating the intent to initiate the state transition, and the exoskeleton system 100 can become more sensitive to the state transition in anticipation of the user 101 making the intended state transition. Such increased sensitivity may be desirable to prevent false negatives or failures to identify a state transition initiated by the user 101.

[0054] Also, providing the user 101 with a single input indicating intent to make a state transition may be desirable because it makes operation of such an exoskeleton system 100 much simpler and more user-friendly than a fully supervised system having multiple buttons mapped to different specific state transitions or systems, where a single button is mapped to fewer than all state transitions (e.g., as shown in FIGS. 7-9). Providing the user 101 with a single input indicating intent to make a state transition may be desirable over unsupervised methods because providing the user 101 with the ability to indicate intent to make a state transition helps prevent false positives and false negatives for state transitions by providing variable sensitivity to state transitions based on the user's intent or desire to make a state transition, which may be associated with an increased likelihood that the state transition will occur.

[0055] To further illustrate the differences between the fully supervised intent recognition methods of Figures 7-9 and the semi-supervised method of Figure 11, it may be useful to focus on an example in which a user has multiple options for making a state transition from a given state. For example, as shown in state diagram 1100 of Figure 11, a user in standing state 625 has the option of transitioning to sitting state 605 via sitting maneuver state 635 or to walking state 650. As shown in the example of Figure 11, if user 101 presses button 1130, user 101 has the option of initiating standing-to-sitting transition 630 or standing-to-walk transition 645, and exoskeleton system 100 can be more sensitive to both potential transitions 630, 645 and can respond to user 101 initiating either of the potential transitions 630, 645.

[0056] In contrast, as shown in the examples of FIGS. 7-9 , when the A button 710 is pressed, the user 101 is forced to transition from standing to walking 645, or at the very least, does not have the option of transitioning from standing to sitting 630, which is an unavailable action. Thus, a fully supervised approach may limit the user 101's mobility options, while a semi-supervised approach (e.g., as shown in FIG. 11 ) may allow the user to indicate their intent to make a state transition without explicitly or implicitly specifying one or more particular state transitions. Stated another way, a fully supervised approach may limit the user 101's mobility options, while the semi-supervised approach of various embodiments does not limit the user 101's mobility options, allowing the exoskeleton system 100 to adapt to the user's 101's movements without restriction.

[0057] The difference between fully supervised and semi-supervised intent recognition can also be illustrated when examining state machines in which a state has a larger number of possible state transitions. For example, Figure 12 illustrates an example state machine 1200 in a fully supervised intent recognition method 1201 in which the standing state 625 has eight possible transitions 630, 645, 1205, 1215, 1225, 1235, 1245, 1255 to eight different system states 635, 650, 1210, 1220, 1230, 1240, 1250, 1260.

[0058] More specifically, the user 101 of the exoskeleton system 100 has the option to transition from a standing state 625 to a sitting state 635 via a standing to sitting transition 630, to a walking state 650 via a standing to walking transition 645, to a jumping state 1210 via a standing to jumping transition 1205, to a lunging state 1220 via a standing to lunging transition 1215, to a crouching state 1230 via a standing to crouching transition 1225, to a diving state 1240 via a standing to diving transition 1235, to a sprinting state 1250 via a standing to sprinting transition 1245, and to a jogging state 1260 via a standing to jogging transition 1255.

[0059] 12, the user interface 515 may have an A button 710 mapped to the standing to sitting transition 630. When the A button 710 is pressed in this example, the exoskeleton system 100 may begin transitioning through the standing to sitting transition 630 to the sitting state 635, with no other states and transitions available when the A button is pressed.

[0060] 13 illustrates a state machine 1200 in a supervised intent recognition method 1300 in which the standing state 625 has eight possible transitions to eight respective states, and four buttons 710, 720, 740, 750 are each mapped to a single transition and state pair. More specifically, the A button 710 is mapped to the standing to sitting transition 630, the B button 720 is mapped to the standing to jump transition 1205, the C button 740 is mapped to the standing to lunge transition 1215, and the D button 750 is mapped to the standing to crouch transition 1225.

[0061] Similar to the example of FIG. 12 , the method 1300 of FIG. 13 illustrates that when each of the buttons 710, 720, 740, and 750 is pressed, it triggers a transition to the state to which the given button is mapped, and does not make other transitions or states available. In this example, other state transitions are only available upon pressing their respective associated buttons from the original state. While the example of FIG. 13 illustrates only four buttons mapped to four respective state and transition pairs, in further embodiments, each of the states can be mapped to a respective button. In other words, for the example state machine 1200 of FIGS. 12 and 13 in further embodiments, each of the eight state-transition pairs can be mapped to a respective button. Thus, if the user 101 wishes to transition from the standing state 625 to another state, the user 101 needs to press the specific button associated with the given state or state transition to initiate the transition to the desired state.

[0062] In contrast, Figure 14 illustrates an example of a semi-supervised intent recognition method 1400 having a state machine 1200 as shown in Figures 12 and 13, and a user interface 515 having a single button 730 for indicating an intent to make a state transition. As shown in Figure 14, the user 101 can be in the standing state 625 and can press the X button 730 to indicate an intent or desire to make a state transition, and the exoskeleton system 100 can be more sensitive to identifying state transitions, allowing the exoskeleton system 100 to initiate any of the eight possible state transitions shown in the example of Figure 14 based on whether a state transition is detected from the user's behavior, or alternatively, choose not to initiate a state transition if none is detected.

[0063] In other words, in the semi-supervised intent recognition method 1400 of FIG. 14, all possible state transitions remain possible only because the manual input (X button 730) indicates that the exoskeleton system 100 will be more sensitive to detecting any possible transitions from the current state (e.g., by lowering the transition threshold to possible actions).

[0064] Nor is the transition possible, nor is the user 101 forced or required to make a state transition. However, in the fully supervised example of Figure 13, if the B button 720 is pressed and the current standing configuration state 625 is deemed stable for the user 101 to transition, the exoskeleton system 100 initiates the standing to jumping transition 1205. In the example of Figure 14, if the X button 730 is pressed and the user 101 has not done anything to indicate that the transition should, is about to, or is occurring, the transition does not occur.

[0065] Additionally, while various embodiments of the semi-supervised intent recognition method are discussed having a single button (e.g., X button 730), it should be clear that various embodiments can include a single input type with one or more input methods for the single input type. For example, in some embodiments, the exoskeleton system 100 can include first and second X buttons 730 located on the left and right actuator units 110A, 110B of the exoskeleton system 100, respectively, and the user 101 can press either of the buttons 730 to cause the exoskeleton system 100 to be more sensitive or responsive to identifying state transitions. Also, a single input type can be associated with multiple input methods in some embodiments. For example, the user 101 can press the X button 730, tap the body of the exoskeleton system 100, or provide a voice command to cause the exoskeleton system 100 to be more sensitive or responsive to identifying state transitions.

[0066] One way to mathematically explain the difference between fully supervised and semi-supervised methods is to examine the probabilities of possible state transitions from a given starting state. In fully supervised methods for various state machines (e.g., state machine 600 of FIG. 6), the probability of transitioning from standing 625 to walking 650 can be equal to N (i.e., P(walking / standing) = N). The probability of transitioning from standing 625 to standing 625 is then 1-N (i.e., P(standing / standing) = 1-N), in which case exoskeleton system 100 does not allow the transition from standing to walking to occur (e.g., due to stability features). In various fully supervised methods, the probability of transitioning from standing 625 to sitting 635 is equal to 0 (i.e., P(sit / standing) = 0), because manual input can map only a single desired transition from a single starting state.

[0067] In a semi-supervised method for such an identical state machine (e.g., state machine 600 of FIG. 6 ), the probability of transitioning from standing 625 to walking 650 can be equal to A (i.e., P(walking / standing) = A). The probability of transitioning from standing 625 to standing 625 is B (i.e., P(standing / standing) = B). The probability of transitioning from standing 625 to sitting 635 is 1−AB (i.e., P(sit / standing) = 1−AB). This may be because, in some embodiments of the semi-supervised intent recognition method, the exoskeleton system 100 remains interpretive of the desired state transition from a given starting state, allowing the exoskeleton system 100 to decide between sitting 635, walking 650, or remaining standing 625.

[0068] 15, various examples illustrate a semi-supervised intent recognition method 1500 according to one embodiment that can be implemented by the exoskeleton device 510 (see FIG. 5) of the exoskeleton system 100. The method 1500 begins at 1505, where the exoskeleton system 100 operates in a first mode with sensitivity to detect state transitions at a first sensitivity level. At 1510, a determination is made whether a state transition is identified, and if so, at 1515, the exoskeleton device facilitates the identified state transition, and the method 1500 repeats, returning to 1505, where the exoskeleton system 100 operates in the first mode with sensitivity to detect state transitions at the first sensitivity level. However, if at 1510 a state transition is not identified, then at 1520 a determination is made whether a state transition intent input is received, and if not, the method 1500 repeats returning to 1505 where the exoskeleton system 100 operates in the first mode with sensitivity to detect state transitions at a first sensitivity level.

[0069] For example, the exoskeleton 100 may operate in a standard sensitivity mode (e.g., a first mode) and may identify one or more state transitions initiated or made by the user 101, and may therefore operate to support the user as needed for such identified one or more state transitions. The exoskeleton system 100 may also monitor or wait for a state transition intent input to be received, which may be received in various suitable manners, such as via pressing a button on the user interface 515, via tactile input, or via audio input, as discussed herein.

[0070] In various embodiments, the exoskeleton system 100 can operate and transition the user 101 through some or all available states during a given operating session without ever receiving a state transition intent input. For example, the exoskeleton system 100 can be powered up, operate in various positional states, and then powered off without ever receiving a state transition intent input. In other words, in various embodiments, the exoskeleton system 100 can become fully functional and have the ability to move through all available positional states and transitions without ever receiving a state transition intent input.

[0071] Returning to method 1500, if a state transition intent input is received at 1520, method 1500 then proceeds to 1525 where the exoskeleton system 100 operates in a second mode with sensitivity to detect state transitions at a second sensitivity level. At 1530, a determination is made whether a state transition is identified, and if so, at 1535, the exoskeleton system 100 facilitates the identified state transition and method 1500 repeats back to 1525 where the exoskeleton system 100 operates in a second mode with sensitivity to detect state transitions at the second sensitivity level.

[0072] However, if at 1530 a state transition is not identified, the method 1500 proceeds to 1540 where a determination is made whether a second mode timeout has occurred. If not, the method 1500 repeats back to 1525 where the exoskeleton system 100 operates in the second mode with sensitivity to detect state transitions at the second sensitivity level. However, if a second mode timeout is determined, the method 1500 then repeats back to 1505 where the exoskeleton system 100 operates in the first mode with sensitivity to detect state transitions at the first sensitivity level.

[0073] For example, when a state transition intent input is received by the exoskeleton system 100, the exoskeleton system 100 can switch from detecting state transitions at a first sensitivity level in a first mode to detecting state transitions at a second sensitivity level in a second mode, where the first and second sensitivity levels are different. The exoskeleton system 100 can monitor state transitions and facilitate one or more identified state transitions until a timeout for operating in the second mode occurs. However, it is not necessary that a state transition be ever identified and / or facilitated while operating in the second mode before the second mode timeout occurs.

[0074] As discussed herein, in various examples, the second sensitivity level of the second mode can be more sensitive to detecting or identifying state transitions compared to the first sensitivity level of the first mode. The greater sensitivity of the second sensitivity level can be achieved in various suitable manners, including by lowering one or more thresholds associated with identifying one or more state transitions or by removing or modifying criteria for identifying one or more state transitions. However, in various embodiments, the thresholds of the set of criteria and / or a subset of criteria need not be changed, removed, or modified. Also, in some embodiments, one or more thresholds can be increased if the overall effect of the difference between the second sensitivity level and the first sensitivity level results in a greater overall sensitivity of the second sensitivity level. In further embodiments, the first and second modes can differ in any suitable manner such that, for some sets of sensor data, a given state transition is not identified as present when the exoskeleton system 100 is operating in the first mode but is identified as present when the exoskeleton system 100 is operating in the second mode.

[0075] The second mode timeout can be generated or implemented in a variety of suitable manners. In some embodiments, the second mode timeout can comprise a timer corresponding to the time a given second mode session has been active (e.g., the amount of time from when a switch from the first mode to the second mode occurs), and the second mode timeout can occur when the timer reaches or exceeds a defined timeout threshold. For example, the timeout threshold can be a number of seconds or minutes, including 1 second, 5 seconds, 10 seconds, 20 seconds, 30 seconds, 45 seconds, 60 seconds, 90 seconds, 2 minutes, 3 minutes, or 5 minutes, etc.

[0076] Such timeout thresholds may be static or variable. In some embodiments, the second mode session may last a defined amount of time. In further embodiments, the second mode session may last a defined amount of time by default, but may be extended or shortened based on any suitable criteria, conditions, sensor data, or the like. For example, in some embodiments, the second mode session may terminate after a state transition is identified and / or the identified state transition is facilitated.

[0077] The intent recognition method can be used in a variety of suitable applications. One example embodiment includes an intent recognition method for a lower limb exoskeleton system 100 for supporting community mobility for seniors. The exoskeleton system 100 can be designed to assist with transitions between seated and standing positions, climbing stairs, and providing assistance during walking maneuvers. In this example, the user is provided with a single input to the exoskeleton system 100 in the form of two knocks or taps on the exterior of the exoskeleton system 100. This manual interaction by the user 101 can be detected through monitoring the exoskeleton system 100's integrated accelerometer or other sensor 513. The exoskeleton system 100 can interpret the input from the user 101 as an indication that a change in behavior is occurring. The exoskeleton system 100 can utilize an unsupervised intent recognition method that monitors the device sensors 513 and observes changes in the user's behavior to identify intent, although the particular method can be tuned to be very conservative to avoid false indications of intent. When intent is indicated by the user 101, the required confidence threshold for the method can be lowered, allowing the exoskeleton system 100 to be much more sensitive to interpreting what constitutes a triggered movement and to respond proactively.

[0078] In such an example, the subject may be donning the exoskeleton system 100 from a seated position, and the only available state transition to the device is then to stand up. When the user 101 taps the exoskeleton system 100 twice, the exoskeleton system 100 may relax the threshold requirement for standing for a fixed period of time, which may be set to 5 seconds for purposes of this example. If the user 101 is not attempting to initiate a standing action, the intent indication simply times out and returns the conservative threshold. If the user 101 is attempting to initiate a standing action, the exoskeleton system 100 captures the movement and responds with assistance accordingly. Once in a standing position, the user 101 may perform a variety of actions, including walking, transitioning to sitting, climbing stairs, or descending stairs. In this case, the user 101 may decide not to tap the machine and instead begin walking. At this point, the device may still be able to respond to the action, but may require much more reliable identification of the targeted action.

[0079] After stopping walking, the user 101 intends to climb stairs. The user 101 taps the device twice to indicate an upcoming change in intended behavior, and then begins completing the movement. Here, the user's indicated intent does not specify to the exoskeleton system 100 which behavior the user 101 intends to transition to, but simply that the transition is likely to occur in the near future. The exoskeleton system 100 observes that the user 101 is standing, and using a more sensitive transition threshold, the exoskeleton system 100 allows certain transitions to behavioral modes to occur.

[0080] Embodiments of the present disclosure can be described in light of the following clauses. 1. A wearable pneumatic exoskeleton system configured to run a semi-supervised intent recognition control program, the exoskeleton system comprising: left and right pneumatic leg actuator units configured to be associated with the left and right legs, respectively, of a user wearing the exoskeleton system, and configured to estimate and transition between a plurality of physical states including at least a standing state, a sitting state, and a walking state, each of the left and right pneumatic actuator units comprising: a rotatable joint configured to be aligned with a rotation axis of a knee of the user wearing the exoskeleton system; an upper arm coupled to the rotatable joint and extending along the length of the thigh above the knee of the user wearing the exoskeleton system; a forearm coupled to the rotatable joint and extending along the length of the lower leg below the knee of the user wearing the exoskeleton system; the left and right pneumatic leg actuator units including an inflatable bellows actuator configured to actuate the upper and lower arms by extending a length of the bellows actuator when pneumatically inflated by introducing air fluid into a bellows cavity; a pneumatic system configured to introduce air fluid into the bellows actuators of the pneumatic leg actuator units to independently operate the bellows actuators; 1. An exoskeleton computing device, comprising: A plurality of sensors; A user input having a state transition intention input button; a memory storing at least a semi-supervised intention recognition control program; a processor that executes the semi-supervised intention recognition control program to control the pneumatic system based at least in part on data acquired by the exoskeleton computing device, the data including sensor data acquired from the plurality of sensors; and by executing the semi-supervised intention recognition and control program, the exoskeleton system operating in a first mode with sensitivity to identify state transitions of the exoskeleton system at a first sensitivity level; identifying a first state transition while operating in the first mode and using the first sensitivity level, and responsively facilitating the identified first state transition by actuating the exoskeleton system; receiving a state transition intent input by a user pressing the state transition intent input button, and in response modifying the exoskeleton system to operate in a second mode with sensitivity to detect state transitions at a second sensitivity level that is more sensitive than the first sensitivity level; identifying a second state transition while operating in the second mode and using the second sensitivity level, and responsively facilitating the identified second state transition by actuating the exoskeleton system; in response to a timeout of the second mode, switching back to operating in the first mode and using the first sensitivity level. The wearable pneumatic exoskeleton system. 2. The wearable pneumatic exoskeleton system described in clause 1, wherein identifying the second state transition at the second sensitivity level is based at least in part on a set of sensor data obtained from at least some of the plurality of sensors of the exoskeleton computing device, and the set of sensor data identifies the second state transition at the second sensitivity level but does not identify the second state transition at the first sensitivity level of the first mode due to the first sensitivity level being less sensitive than the second sensitivity level. 3. By executing the semi-supervised intention recognition and control program, the exoskeleton system further receiving a second state transition intent input after switching back to operating in the first mode, and in response thereto, modifying the exoskeleton system to operate in the second mode with a sensitivity to detect state transitions at the second sensitivity level that is more sensitive than the first sensitivity level; and in response to a further second mode timeout, switching back to operating in the first mode and using the first sensitivity level, wherein said switching back to operating in the first mode is performed by the exoskeleton system without identifying or facilitating a state transition while operating in the second mode after receiving the second state transition intent input. 3. A wearable pneumatic exoskeleton system according to clause 1 or 2. 4. Identifying the second state transition while operating in the second mode and using the second sensitivity level includes: determining that the exoskeleton is in a standing state, wherein the exoskeleton has the option to transition from the standing state to either a sitting state or a walking state while operating in the second mode and after receiving the state transition intent input; monitoring state transitions, including state transitions to either the sitting state or the walking state; identifying the second state transition using the second sensitivity level, the second state transition being one of the sitting state or the walking state, and in response, activating the exoskeleton system to facilitate the transition to one of the sitting state or the walking state; 4. The wearable pneumatic exoskeleton system of any one of clauses 1 to 3, comprising: 5. A wearable pneumatic exoskeleton system described in any of clauses 1 to 4, wherein receiving a state transition intent input does not trigger a state transition by the exoskeleton system, and receiving a state transition intent input does not limit the feasible physical state transition options of the exoskeleton system. 6. An exoskeleton system configured to run a semi-supervised intent recognition and control program, comprising: An actuator unit, a joint configured to be aligned with a knee of a user wearing the leg actuator unit; an upper arm coupled to the joint and extending along the length of the thigh above the knee of the user wearing the leg actuator unit; a forearm coupled to the joint and extending along the length of the lower leg below the knee of the user wearing the leg actuator unit; and an actuator configured to actuate the upper arm and forearm and move the leg actuator unit into a plurality of different positional states; By executing the semi-supervised intention recognition and control program, the exoskeleton system: receiving a state transition intent input and, in response thereto, changing the exoskeleton system from operating in a first mode with a sensitivity to detect state transitions at a first sensitivity level to operating in a second mode with a sensitivity to detect state transitions at a second sensitivity level that is more sensitive than the first sensitivity level; operating in the second mode and identifying a second state transition while using the second sensitivity level, and responsively facilitating the identified state transition by actuating the exoskeleton system. Exoskeleton system. 7. The exoskeleton system of clause 6, wherein by executing the semi-supervised intention recognition control program, the exoskeleton system further switches back to operating in the first mode and using the first sensitivity level for identifying state transitions in response to a second mode timeout. 8. An exoskeleton system as described in clause 6 or 7, wherein identifying the state transition at the second sensitivity level is based at least in part on a set of sensor data obtained from one or more sensors of the exoskeleton system, the set of sensor data identifying the state transition at the second sensitivity level but not identifying the state transition at the first sensitivity level of the first mode. 9. Identifying the state transition while operating in the second mode and using the second sensitivity level includes: determining that the exoskeleton system is in a standing state, wherein the exoskeleton system has the option to transition from the standing state to either a sitting state or a walking state while operating in the second mode and after receiving the state transition intent input; monitoring state transitions, including state transitions to either the sitting state or the walking state; identifying the state transition using the second sensitivity level, the state transition being one of the sitting state or the walking state, and in response, facilitating the transition to one of the sitting state or the walking state by actuating the exoskeleton system; 9. The exoskeleton system of any of clauses 6 to 8, comprising: 10. An exoskeleton system described in any of clauses 6 to 9, wherein receiving a state transition intent input does not trigger a state transition by the exoskeleton system, and receiving a state transition intent input does not limit the state transition options of the exoskeleton system. 11. A computer-implemented method of semi-supervised intent recognition for an exoskeleton system, comprising: In response to a state transition intent input, changing the exoskeleton system from operating in a first mode with a sensitivity to detect state transitions at a first sensitivity level to operating in a second mode with a sensitivity to detect state transitions at a second sensitivity level that is more sensitive than the first sensitivity level; identifying a state transition while operating in the second mode and using the second sensitivity level; facilitating the identified state transition by actuating the exoskeleton system; and The computer-implemented method comprising: 12. The computer-implemented method of clause 11, further comprising, in response to a second mode timeout, switching to operating in the first mode and using the first sensitivity level. 13. The computer-implemented method of clause 11 or 12, wherein identifying the state transition at the second sensitivity level is based at least in part on sensor data that identifies the state transition at the second sensitivity level but does not identify the state transition at the first sensitivity level of the first mode. 14. Identifying the state transition while operating in the second mode and using the second sensitivity level includes: determining that the exoskeleton system is in a first physical state, the exoskeleton system having options to transition from the first physical state to a plurality of physical states while operating in the second mode and after receiving the state transition intent input; identifying the state transition using the second sensitivity level, the state transition being one of the plurality of physical states, and responsively activating the exoskeleton system to facilitate the transition to the one of the plurality of physical states; 14. The computer-implemented method of any of clauses 11 to 13, comprising: 15. The computer-implemented method of clause 14, wherein the first physical state is a standing state, one of the plurality of available physical state transitions is a transition to a sitting state, and another of the plurality of physical state transitions is a transition to a walking state. 16. The computer-implemented method of any of clauses 11 to 15, wherein receiving a state transition intent input does not trigger a state transition by the exoskeleton system. 17. The computer-implemented method of any of clauses 11 to 16, wherein receiving a state transition intent input does not limit the state transition options of the exoskeleton system. 18. Operating the exoskeleton system in the first mode with a sensitivity that identifies state transitions of the exoskeleton system at the first sensitivity level; identifying a second state transition while operating in the first mode and using the first sensitivity level, and responsively facilitating the identified second state transition by actuating the exoskeleton system; 12. The computer-implemented method of claim 11, further comprising: 19. receiving a second state transition intent input after switching back to operating in the first mode from the second mode; modifying the exoskeleton system to operate in a second mode with sensitivity to identify state transitions at the second sensitivity level in response to the second state transition intent input; and in response to a further second mode timeout, switching again to operating in the first mode and using the first sensitivity level, wherein the switching again to operating in the first mode is performed by the exoskeleton system without identifying or facilitating a state transition while operating in the second mode and after receiving the second state transition intent input. 12. The computer-implemented method of claim 11, further comprising:

[0081] The described embodiments are susceptible to various modifications and alternative forms, specific embodiments of which have been shown by way of example in the drawings and are herein described in detail. It is to be understood, however, that the described embodiments are not limited to the specific forms or manners disclosed, but on the contrary, the disclosure covers all modifications, equivalents, and alternatives.

Claims

1. 1. An exoskeleton system configured to execute a semi-supervised intention recognition control program, the exoskeleton system having leg actuator units; This leg actuator unit is a joint configured to align with a knee of a user wearing the leg actuator unit; an upper arm coupled to the joint and extending along the length of the thigh above the knee of the user wearing the leg actuator unit; a forearm coupled to the joint and extending along the length of the lower leg below the knee of the user wearing the leg actuator unit; an actuator configured to operate the upper arm and the forearm and move the leg actuator unit into a plurality of different positional states; Executing the semi-supervised intention recognition and control program to the exoskeleton system, receiving a state transition intent input via a user interface having a single button for indicating an intent to cause a state transition, the button indicating only the user's intent to cause a state transition from a first state to a second state selectable from a plurality of possible states, without indicating any particular state transition among a plurality of possible state transitions to be made, and no other interface elements of the user interface can be used to generate the state transition intent input; in response to the state transition intent input, temporarily changing the exoskeleton system from a first mode of operation having sensitivity to detect state transitions at a first sensitivity level to a second mode of operation having sensitivity to detect state transitions at a second sensitivity level that is more sensitive than the first sensitivity level; identifying a state transition among the plurality of possible state transitions to be made while operating in the second mode and using the second sensitivity level, and responsively operating the exoskeleton system to facilitate the identified state transition, wherein the identification of the state transition at the second sensitivity level is based at least in part on a set of sensor data acquired by one or more sensors of the exoskeleton system, the set of sensor data identifying a state transition at the second sensitivity level but not at the first sensitivity level of the first mode; 10. An exoskeleton system, comprising: an exoskeleton configured to automatically switch from the second mode back to the first mode without user input based, at least in part, on a second mode timeout.

2. The first state is a standing state, and the plurality of possible different states, which are the second state, are: Sitting position, Walking state The exoskeleton system of claim 1 , wherein the exoskeleton system can be selected from:

3. The plurality of possible state transitions to be made include: Standing to sitting, and From standing to walking The exoskeleton system of claim 1 ,

4. Identifying a state transition while operating in the second mode and using the second sensitivity level comprises: determining that, after operating in the second mode and receiving the state transition intent input, the exoskeleton system has an option to transition from a standing state to a sitting state or a walking state, and that the exoskeleton system is in the standing state; monitoring state transitions, including state transitions to the sitting state or the walking state; and The exoskeleton system of claim 1 , further comprising using the second sensitivity level to identify a state transition, the state transition being to the sitting state or the walking state, and in response thereto, operating the exoskeleton system to facilitate the transition to the sitting state or the walking state.

5. The exoskeleton system of claim 1 , wherein receiving a state transition intent input does not trigger a state transition of the exoskeleton system, and receiving a state transition intent input does not limit state transition options of the exoskeleton system.

6. An exoskeleton-type system having a semi-supervised intention recognition control program and configured to execute the semi-supervised intention recognition control program, wherein the semi-supervised intention recognition control program is configured to provide the exoskeleton-type system with: in response to a state transition intent input, causing the exoskeleton system to temporarily change from a first mode of operation having sensitivity to detect state transitions at a first sensitivity level to a second mode of operation having sensitivity to detect state transitions at a second sensitivity level that is more sensitive than the first sensitivity level, the state transition intent input being generated via a user interface having a single button for indicating intent to cause a state transition without indicating any particular state transition among a plurality of possible state transitions to be made, and no interface elements other than the user interface may be used to generate the state transition intent input; identifying one state transition of the plurality of possible state transitions to be made while operating in the second mode and using the second sensitivity level, the identification of the state transition at the second sensitivity level being based at least in part on a set of sensor data acquired by one or more sensors of the exoskeleton system, the set of sensor data identifying a state transition at the second sensitivity level but not at the first sensitivity level of the first mode; An exoskeleton system, characterized in that the identified state transition is promoted by operating the exoskeleton system.

7. 7. The exoskeleton system of claim 6, wherein the state transition intent input indicates only an intent to cause a state transition from a first state to a second state, the second state being selectable from a plurality of different possible states.

8. The first state is a standing state, and the plurality of possible different states, which are the second state, are: Sitting position, Walking state The exoskeleton system of claim 7, wherein the exoskeleton system can be selected from:

9. The plurality of possible state transitions to be made include: Standing to sitting, and From standing to walking The exoskeleton system of claim 6, wherein:

10. The exoskeleton system of claim 6 , wherein no other interface elements of the user interface can be used to generate the state transition intent input.

11. 1. An exoskeleton system comprising: and in response to a state transition intent input, causing the exoskeleton system to temporarily change operation from a first mode having sensitivity to detect state transitions at a first sensitivity level to a second mode of operation having sensitivity to detect state transitions at a second sensitivity level greater than the first sensitivity level, the state transition intent input being generated via a user interface having a single interface element for indicating intent to cause a state transition, the single interface element indicating only the user's intent to cause a state transition from a first state to a second state selectable from a plurality of possible states, without indicating any particular state transition among a plurality of possible state transitions to be made; there are no other interface elements of the user interface that can be used to generate the state transition intent input; there are no other interface elements of the user interface that can be used to generate the state transition intent input; 1. An exoskeleton system, comprising: a state transition intent input indicating an intent to perform a state transition from a first state to a second state, the second state being configured to be selected from a plurality of different possible states.

12. The exoskeleton system of claim 11 , wherein the state transition intent input indicates an intent to perform the state transition without indicating any particular state transition among multiple possible state transitions that may be performed.

13. The plurality of possible state transitions to be made include: Standing to sitting, and From standing to walking 13. The exoskeleton system of claim 12, wherein:

14. The first state is a standing state, and the plurality of possible different states, which are the second state, are: Sitting position, and Walking state The exoskeleton system of claim 11, wherein:

15. The exoskeleton system of claim 11 , wherein the state transition intent input is generated by pressing a button on a user interface.

16. 16. The exoskeleton system of claim 15, wherein the state transition intent input can be generated solely by the user interface by pressing a single button on the user interface, and no other interface elements of the user interface other than the single button can be used to generate the state transition intent input.

17. The exoskeleton system further comprises: identifying one state transition of a plurality of possible state transitions to make while operating in the second mode and using the second sensitivity level; The exoskeleton system of claim 11 , wherein the identified state transition is facilitated by manipulating the exoskeleton system.

18. 12. The exoskeleton system of claim 11, wherein identifying a state transition at the second sensitivity level is based at least in part on a set of sensor data acquired by one or more sensors of the exoskeleton system, the set of sensor data identifying a state transition at the second sensitivity level but not identifying a state transition at the first sensitivity level in the first mode.

19. Identifying a state transition while operating in the second mode and using the second sensitivity level includes: determining that the exoskeleton system is in a first physical state, the exoskeleton system operating in the second mode and having options for transitioning from the first physical state to a plurality of physical states after receiving the state transition intent input; 12. The exoskeleton system of claim 11, further comprising using the second sensitivity level to identify the state transition, the state transition being to one of the plurality of physical states, and in response, manipulating the exoskeleton system to facilitate the transition to one of the plurality of physical states.

Images