User feedback system and method

Abstract

A user feedback system for a user of a delivery device in a delivery ecosystem comprises an acquisition processor configured to acquire one or more user factors indicative of a state of the user, and an estimation processor configured to identify a one-stage or two-stage correspondence between the one or more user factors indicative of the acquired user state and at least a first feedback action, the feedback action being expected to change the state of the user as at least partially indicated by the one or more user factors.

Description

[Technical field]

[0001] The present invention relates to a user feedback system and method for a user of a delivery device.

[0002] The "Background" description provided herein is intended to provide a general overview of the background of the present disclosure. The work of the inventors and aspects of the present specification that may not be admitted as prior art at the time of filing to the extent described in this Background section are not admitted explicitly or implicitly as prior art to the present disclosure.

[0003] Aerosol delivery systems have proven popular with users because they allow the active ingredient (such as nicotine) to be delivered to the user conveniently and on demand as needed.

[0004] As an example of an aerosol delivery system, an electronic cigarette (e-cigarette) typically includes a reservoir of liquid feedstock containing a formulation (usually containing nicotine) from which an aerosol is generated, for example, by thermal vaporization. The aerosol source of the aerosol delivery system may thus comprise a heater having a heating element configured to receive the liquid feedstock from the reservoir, for example, by wicking / capillary action. Other feedstocks may also be similarly heated to generate aerosols, such as botanicals or gels containing active ingredients and / or flavorings. Thus, more generally, e-cigarettes are considered to contain or receive a payload for thermal vaporization.

[0005] While a user inhales on the device, power is supplied to the heating element, causing an aerosol source (part of the payload) adjacent to the heating element to vaporize, generating an aerosol that is inhaled by the user. Such devices are typically provided with one or more intake holes located away from the mouthpiece end of the system. When a user inhales on a mouthpiece connected to the mouthpiece end of the system, air is drawn in through the inlet hole and passes through the aerosol source. There is a flow path connecting the aerosol source to the mouthpiece opening, so that the air drawn in and passing through the aerosol source travels along the flow path to the mouthpiece opening, carrying a portion of the aerosol from the aerosol source. The aerosol-carrying air exits the aerosol delivery system through the mouthpiece opening, and is inhaled by the user.

[0006] Typically, electrical current is provided to the heater upon inhalation / puffing by the user on the device. Typically, electrical current is provided to the heater (e.g., a resistive heating element) in response to activation of an airflow sensor along the flow path during inhalation / inhalation / puffing by the user or activation of a button by the user. Heat generated by the heating element is used to vaporize the formulation. The released vapour mixes with air drawn into the device by the puffing consumer to form an aerosol. Alternatively or additionally, the heating element is typically used to heat, rather than burn, botanical matter such as tobacco, causing the active ingredient to be released as a vapour / aerosol.

[0007] The manner in which a user interacts with an e-cigarette (e.g., the amount of vaporized / aerosolized payload consumed by the user and / or the respective usage pattern) and the actual or perceived utility of the interaction are believed to be influenced by the user's state, which may be expressed at least in part in colloquial terms as the respective mood(s) and / or subjective need(s).

[0008] As a result, it would be useful to provide a delivery mechanism that is more responsive to the user's condition. Summary of the Invention

[0009] In a first aspect, there is provided a user feedback system for a user of a delivery device in a delivery ecosystem according to claim 1.

[0010] In another aspect, there is provided a method of user feedback to a user of a delivery device in a delivery ecosystem according to claim 29.

[0011] Other aspects and features of the present invention are defined in the accompanying claims.

[0012] It is to be understood that both the foregoing general summary of the disclosure and the following detailed description are illustrative of the present disclosure but are not restrictive of the present disclosure.

[0013] The present disclosure and many of the attendant advantages thereof will be readily appreciated as they become better understood by reference to the following detailed description, when considered in connection with the accompanying drawings, in which: [Brief description of the drawings]

[0014] [Figure 1] FIG. 2 is a schematic diagram of a delivery device according to an embodiment herein. [Diagram 2] FIG. 2 is a schematic diagram of a body of a delivery device according to an embodiment herein. [Diagram 3] FIG. 2 is a schematic diagram of a cartomizer of a delivery device according to an embodiment herein. [Figure 4] FIG. 2 is a schematic diagram of a body of a delivery device according to an embodiment herein. [Diagram 5] FIG. 1 is a schematic diagram of a delivery ecosystem according to embodiments herein. [Figure 6] FIG. 1 is a schematic diagram of a user feedback system, according to an embodiment herein. [Figure 7] FIG. 1 is a flow diagram of a method for user feedback to a user of a delivery device in a delivery ecosystem according to an embodiment herein. Description of the embodiments

[0015] A user feedback system and method are disclosed. In the following description, many specific details are presented to enable a thorough understanding of the embodiments of the present disclosure. However, it will be apparent to those skilled in the art that the adoption of these specific details is not necessary to realize the embodiments of the present disclosure. On the contrary, specific details that are understood by those skilled in the art are omitted, where necessary, for the sake of clarity.

[0016] As mentioned above, the present disclosure relates to a user feedback system for improving the responsiveness of a delivery device to a user.

[0017] The term "delivery device" may encompass systems that deliver at least one substance to a user, and may include non-combustible aerosol delivery systems that release compounds from an aerosol-generating material without combustion of the aerosol-generating material, such as electronic cigarettes, tobacco heating products, and hybrid systems that generate an aerosol using a combination of aerosol-generating materials, and aerosol-free delivery systems that deliver at least one substance to a user orally, nasally, transdermally, or otherwise without the formation of an aerosol (including, but not limited to, oral products such as lozenges, gums, patches, articles containing inhalable powders, and oral tobacco products including snus or moist snuff) (the at least one substance may or may not contain nicotine).

[0018] The substance to be delivered may be an aerosol-generating material or a material not subject to aerosolization, either of which may optionally contain one or more active constituents, one or more flavorings, one or more aerosol-forming materials, and / or one or more other functional materials.

[0019] Currently, the most common example of such a delivery device is an aerosol delivery system (e.g., a non-flammable aerosol delivery system) or an electronic vapor delivery system (EVPS), such as an e-cigarette. Throughout the following description, the term "e-cigarette" may be used interchangeably with delivery device, unless otherwise stated or otherwise indicated by the context. Similarly, the terms "vapour" and "aerosol" are referred to equivalently in this specification.

[0020] In general, the electronic vapor / aerosol delivery system may be an electronic cigarette, also known as a vaping device or an electronic nicotine delivery system (END), although it is noted that the presence of nicotine in the aerosol-generating (e.g., aerosolizable) material is not a requirement. In some embodiments, the non-combustion aerosol delivery system is a tobacco heating system, also known as a non-combustion heating system. One example of such a system is a tobacco heating system. In some embodiments, the non-combustion aerosol delivery system is a hybrid system that generates aerosol by a combination of aerosol-generating materials, one or more of which may be heated. Each of the aerosol-generating materials may be, for example, in solid, liquid, or gel form, and may or may not contain nicotine. In some embodiments, the hybrid system includes liquid or gel aerosol-generating materials as well as solid aerosol-generating materials. The solid aerosol-generating materials may include, for example, tobacco or non-tobacco products. Meanwhile, in some embodiments, the non-combustion aerosol delivery system generates vapor / aerosol from such one or more aerosol-generating materials.

[0021] Typically, a non-flammable aerosol delivery system may include a non-flammable aerosol delivery device and an article (sometimes referred to as a consumable) for use with the non-flammable aerosol delivery system. However, it is contemplated that an article that itself includes a means for powering an aerosol generating component (e.g., an aerosol generator, such as a heater, a vibrating mesh, etc.) may itself constitute a non-flammable aerosol delivery system. In one embodiment, the non-flammable aerosol delivery device may include a power source and a controller. The power source may be an electrical power source or a heat generating power source. In one embodiment, the heat generating power source includes a carbon substrate that can be energized to provide power in the form of heat to an aerosolizable material or a heat transfer material in proximity to the heat generating power source. In one embodiment, a power source, such as a heat generating power source, is provided in the article to enable non-flammable aerosol delivery. In one embodiment, an article for use with a non-flammable aerosol delivery device may include an aerosolizable material.

[0022] In some embodiments, the aerosol generating component is a heater capable of releasing one or more volatile substances from the aerosolizable material to form an aerosol upon interaction with the aerosolizable material. In one embodiment, the aerosol generating component is capable of generating an aerosol from the aerosolizable material without the application of heat. For example, the aerosol generating component may be capable of generating an aerosol from the aerosolizable material without the application of heat, for example by one or more of vibrational, mechanical, pressurized, or electrostatic means.

[0023] In some embodiments, the aerosolizable material may include an active material, an aerosol-forming material, and optionally one or more functional materials. The active material may include nicotine (optionally contained in tobacco or a tobacco derivative) or one or more other non-olfactory physiologically active materials. A non-olfactory physiologically active material is a material that is included in the aerosolizable material to achieve a physiological response other than the sense of smell. The aerosol-forming material may include one or more of glycerin, glycerol, propylene glycol, diethylene glycol, triethylene glycol, tetraethylene glycol, 1,3-butylene glycol, erythritol, meso-erythritol, ethyl vanillate, ethyl laurate, diethyl sulfate, triethyl citrate, triacetin, diacetin mixture, benzyl benzoate, benzyl phenylacetate, tributyrin, lauryl acetate, lauric acid, myristic acid, and propylene carbonate. The one or more functional ingredients may include one or more of a fragrance, a carrier, a pH adjuster, a stabilizer, and / or an antioxidant.

[0024] In some embodiments, an article for use with a non-flammable aerosol delivery device may include an aerosolizable material or an area for receiving an aerosolizable material. In one embodiment, an article for use with a non-flammable aerosol delivery device may include a mouthpiece. The area for receiving an aerosolizable material may be a storage area that stores the aerosolizable material. For example, the storage area may be a reservoir. In one embodiment, the area for receiving an aerosolizable material may be separate from the aerosol generation area or may be coupled to the aerosol generation area.

[0025] As an alternative or in addition to an aerosol delivery system, the delivery device may include any device that introduces / allows for the introduction of an active ingredient into the user's body so that the active ingredient can be effective.

[0026] Thus, exemplary delivery devices include devices that, for example, dispense an aerosol into a receptacle, which the user can then remove from the device to inhale or smoke the aerosol, and thus do not necessarily require the direct involvement of the user at the point of consumption.

[0027] In this regard, alternatively or in addition to the above, the delivery device may provide reminders or usage regimes for the user (e.g., reminding the user when to use a snus pouch or other active deliverable, such as a pill), and the delivery device may optionally store and dispense such consumables in accordance with the reminders or usage regimes.

[0028] Similarly, an exemplary delivery device may be a home refill station that mixes e-liquid ingredients for a user and uses the mixture to fill the reservoir of an e-cigarette, thereby determining the type, blend, and / or concentration of active ingredients (all things being equal) that the user consumes. Such a home refill station may be referred to as a "dock," a power charging station, or a device that combines both functions.

[0029] In this regard, a delivery device operating as a vending machine can similarly provide consumable refills or disposable devices based on a mixture and / or selection of e-liquid components, mixed on demand or equivalently selected from a variety of premade mixtures. Similarly, in other embodiments, the vending machine may provide oral products (e.g., snus, snuff, gums, gels, sprays, and other delivery systems such as patches) or other consumable products, including, for example, active ingredients and / or flavorings.

[0030] In each case, the delivery device is operable to affect one or more of the amount, timing, type, blend, and / or concentration of the active ingredient(s) consumed by the user.

[0031] More generally, therefore, the delivery device is operable to affect the properties of the active ingredient that is consumed by the user.

[0032] Of course, multiple delivery devices may operate in tandem to affect this. For example, a home refill station or vending machine may actually operate in conjunction with an e-cigarette to provide active ingredient changes or other feedback to the user. Similarly, a mobile phone may operate in parallel with the e-cigarette to provide information or analysis regarding said changes or other feedback.

[0033] In this sense, a delivery device may actually be a delivery system with multiple devices operating in sequence and / or in parallel to exert the desired influence / feedback, and therefore references herein to a delivery device or a delivery system are to be considered interchangeable unless otherwise stated.

[0034] Referring now to the drawings, in which like reference numbers represent the same or corresponding parts throughout the several views, FIG. 1 is a schematic diagram (not to scale) of a vapor / aerosol delivery system, such as an e-cigarette 10, providing one non-limiting example of a delivery device according to some embodiments of the present disclosure.

[0035] The e-cigarette has a generally cylindrical shape extending along a longitudinal axis indicated by dashed line LA and comprises two main components, namely, a body 20 and a cartomizer 30. The cartomizer comprises an internal chamber containing a reservoir of payload, e.g., a liquid including nicotine, a vaporizer (e.g., a heater), and a mouthpiece 35. It will be understood that, hereinafter, references to "nicotine" are by way of example only and can be replaced with any suitable active ingredient. It will be understood that, hereinafter, references to "liquid" as the payload are by way of example only and can be replaced with any suitable payload, such as botanical matter (e.g., tobacco that is heated rather than burned) or a gel that includes an active ingredient and / or flavoring. The reservoir may be a foam matrix or any other structure that holds the liquid until it is required for delivery to the vaporizer. In the case of liquid / flowable payloads, the vaporizer is for vaporizing the liquid, and the cartomizer 30 may further include a wick or similar mechanism for transporting a small amount of liquid from a reservoir to a vaporization location on or adjacent to the vaporizer. In the following, a heater is used as an example of a vaporizer, although it will be appreciated that other forms of vaporizers (e.g., ultrasonic vaporizers) can also be used, and it will be appreciated that the type of vaporizer used may also depend on the type of payload being vaporized.

[0036] The main body 20 includes a rechargeable cell or battery that powers the e-cigarette 10 and a wiring board that controls the entire e-cigarette. The heater receives power from the battery and, when controlled by the wiring board, vaporizes the liquid. This vapor is then inhaled by the user through the mouthpiece 35. In some specific embodiments, the main body is further provided with a manual activation device 265 (e.g., a button, switch, or touch sensor located on the outside of the main body).

[0037] Although the main body 20 and the cartomizer 30 may be detachable from each other by separation in a direction parallel to the longitudinal axis LA as shown in Fig. 1, when the device 10 is in use, the main body 20 and the cartomizer 30 are mechanically and electrically connected by integral joining through connections shown diagrammatically as 25A and 25B in Fig. 1. The electrical connector 25B of the main body 20 used for connecting to the cartomizer 30 also functions as a socket for connecting a charging device (not shown) when the main body 20 is detached from the cartomizer 30. The battery in the main body 20 of the e-cigarette 10 can be charged by plugging the other end of the charging device into a USB socket. In other embodiments, a cable may be provided for directly connecting the electrical connector 25B of the main body 20 to the USB socket.

[0038] The e-cigarette 10 is provided with one or more holes (not shown in FIG. 1) for air intake. These holes lead to an air passageway through the e-cigarette 10 and into the mouthpiece 35. When the user draws on the mouthpiece 35, air is drawn into this air passageway through one or more air intake holes, preferably located on the exterior of the e-cigarette. When the heater is activated to vaporize the nicotine from the cartridge, the airflow passes through and combines with the resulting vapor, and the combination of airflow and resulting vapor then exits the mouthpiece 35 for the user to draw on. Except for disposable devices, the cartomizer 30 may be removed from the body 20 and discarded (and replaced with another cartomizer, if desired) when the liquid supply is exhausted.

[0039] It should be understood that the e-cigarette 10 shown in Figure 1 is provided by way of example only, and various other implementations are possible. For example, in some embodiments, the cartomizer 30 is provided as two separate components: a cartridge with a liquid reservoir and a mouthpiece (which can be replaced when the reservoir is depleted of liquid), and a vaporizer with a (typically retained) heater. In another example, the charging mechanism may be connected to an additional or alternative power source, such as a car cigarette lighter socket.

[0040] Figure 2 is a schematic (simplified) diagram of the body 20 of the e-cigarette 10 of Figure 1, according to some embodiments of the present disclosure. Figure 2 can generally be considered as a cross-section in a plane passing through the longitudinal axis LA of the e-cigarette 10. Note that various components and details of the body (e.g., wiring and more complex moldings, etc.) have been omitted from Figure 2 for clarity.

[0041] The main body 20 includes a battery or cell 210 that powers the e-cigarette 10 in response to activation of the device by a user. The main body 20 also includes a control unit (not shown in FIG. 2) (e.g., a chip such as an application specific integrated circuit (ASIC) or a microcontroller) that controls the e-cigarette 10. The microcontroller or ASIC includes a CPU or microprocessor. The operation of electronic components such as the CPU is typically controlled, at least in part, by a software program running on the CPU (or other components). Such software programs may be stored in a non-volatile memory such as a ROM that may be embedded in the microcontroller itself or provided as a separate component. The CPU may access the ROM as needed to load and execute individual software programs. The microcontroller also includes appropriate communication interfaces (and control software) for communicating with other devices in the main body 10 as needed.

[0042] The main body 20 further comprises a cap 225 that seals and protects the distal end of the e-cigarette 10. Typically, an air inlet is provided in or adjacent to the cap 225 to allow air to enter the main body 20 when the user draws on the mouthpiece 35. The control unit or ASIC may be located to the side or at one end of the battery 210. In some embodiments, the ASIC is attached to a sensor unit 215 to detect a draw on the mouthpiece 35 (or alternatively, the sensor unit 215 may be provided on the ASIC itself). An air path is provided from the air inlet through the e-cigarette, through the airflow sensor 215 and the heater (of the vaporizer or cartomizer 30) to the mouthpiece 35. Thus, when a user draws on the mouthpiece of the e-cigarette, the CPU detects such a draw based on information from the airflow sensor 215.

[0043] At the end of the body 20 opposite the cap 225 is a connector 25B for joining the body 20 to the cartomizer 30. The connector 25B provides a mechanical and electrical connection between the body 20 and the cartomizer 30. The connector 25B includes a body connector 240 that is metallic (in some embodiments, silver plated) and serves as a terminal for electrical connection (positive or negative) to the cartomizer 30. The connector 25B further includes an electrical contact 250 that provides a second terminal for electrical connection to the cartomizer 30 of opposite polarity to the first terminal, i.e., the body connector 240. The electrical contact 250 is mounted to a coil spring 255. When the body 20 is attached to the cartomizer 30, the connector 25A of the cartomizer 30 presses against the electrical contact 250 to compress the coil spring in an axial direction, i.e., parallel to the longitudinal axis LA (collinear direction). Considering the resilience of the spring 255, this compression tends to expand the spring 255, which has the effect of pressing the electrical contact 250 firmly against the connector 25A of the cartomizer 30, thus helping to ensure a good electrical connection between the body 20 and the cartomizer 30. The body connector 240 and the electrical contact 250 are separated by a trestle 260 which is constructed of a non-conductor (such as plastic) and therefore provides good insulation between the two electrical terminals. The trestle 260 is shaped to aid in the mechanical engagement of the connectors 25A and 25B with each other.

[0044] As mentioned above, the button 265, which represents one form of manual activation device 265, may be located on the outer housing of the main body 20. The button 265 may be implemented using any suitable mechanism operable to be manually activated by a user, such as, for example, a mechanical button or switch, a capacitive or resistive touch sensor, etc. It should also be appreciated that the manual activation device 265 may be located on the outer housing of the cartomizer 30 rather than on the outer housing of the main body 20, in which case the manual activation device 265 may be attached to the ASIC via the connections 25A, 25B. The button 265 may also be located at the end of the main body 20 instead of (or in addition to) the cap 225.

[0045] Figure 3 is a schematic diagram of the cartomizer 30 of the e-cigarette 10 of Figure 1, according to some embodiments of the present disclosure. Figure 3 may be generally considered to be a cross-section in a plane passing through the longitudinal axis LA of the e-cigarette 10. Note that various components and details of the cartomizer 30 (e.g., wiring and more complex moldings, etc.) have been omitted from Figure 3 for clarity.

[0046] The cartomizer 30 includes an air passage 355 extending along a central (longitudinal) axis of the cartomizer 30 from the mouthpiece 35 to a connector 25A that connects the cartomizer 30 to the main body 20. A liquid reservoir 360 is provided around the air passage 335. The reservoir 360 may be realized, for example, by providing cotton or foam saturated with liquid. The cartomizer 30 also includes a heater 365 that heats liquid from the reservoir 360 to generate vapour that flows through the air passage 355 and exits the mouthpiece 35 in response to a user drawing on the e-cigarette 10. The heater 365 is powered through lines 366 and 367, which are connected to opposite poles (positive and negative, or vice versa) of the battery 210 of the main body 20 via a connector 25A (the details of the wiring between the wires 366 and 367 and the connector 25A are omitted from FIG. 3).

[0047] The connector 25A includes an inner electrode 375, which may be silver plated or constructed of any other suitable metal or conductive material. When the cartomizer 30 is connected to the body 20, the inner electrode 375 contacts the electrical contact 250 of the body 20 to provide a first electrical path between the cartomizer 30 and the body 20. Notably, when the connectors 25A and 25B are engaged, the inner electrode 375 pushes against the electrical contact 250, compressing the coil spring 255, thereby helping to ensure good electrical contact between the inner electrode 375 and the electrical contact 250.

[0048] The inner electrode 375 is surrounded by an insulating ring 372, which may be made of plastic, rubber, silicone, or any other suitable material. The insulating ring is surrounded by a cartomizer connector 370, which may be silver plated or made of any other suitable metal or conductive material. When the cartomizer 30 is connected to the body 20, the cartomizer connector 370 contacts the body connector 240 of the body 20 to provide a second electrical path between the cartomizer 30 and the body 20. In other words, the inner electrode 375 and the cartomizer connector 370 function as positive and negative terminals (or vice versa) for supplying power from the battery 210 of the body 20 to the heater 365 of the cartomizer 30 via the supply lines 366 and 367 as needed.

[0049] The cartomizer connector 370 is provided with two lugs or tabs 380A, 380B extending in opposite directions away from the longitudinal axis of the e-cigarette 10. These tabs are used to connect the cartomizer 30 to the body 20 by a bayonet fit with the body connector 240. This bayonet fit provides a secure and robust connection between the cartomizer 30 and the body 20, so that the cartomizer and the body are held in a fixed position relative to each other with minimal wobble or deflection, greatly reducing the chance of any accidental separation. At the same time, the bayonet fit allows for easy and quick connection and separation by connecting by rotation after insertion and disconnecting by withdrawal after rotation (in the opposite direction). Of course, in other embodiments, different forms of connection may be used between the body 20 and the cartomizer 30, such as a snap fit or a screw connection.

[0050] 4 is a schematic diagram of certain details of connector 25B at the end of body 20 according to some embodiments of the present disclosure (although for clarity, most of the internal structure of the connector as shown in FIG. 2 (such as trestle 260) has been omitted). In particular, FIG. 4 shows outer housing 201 of body 20 having a generally cylindrical tube form. This outer housing 201 may, for example, have an inner metal tube with an outer covering such as paper. The outer housing 201 may also include a manual activation device 265 (not shown in FIG. 4) for easy access by the user.

[0051] The body connector 240 extends from the outer housing 201 of the body 20. As shown in FIG. 4, the body connector 240 includes two main portions: a shaft portion 241 in the shape of a hollow cylindrical tube sized to fit snugly within the outer housing 201 of the body 20, and a lip portion 242 facing radially outwardly away from the e-cigarette's primary longitudinal axis (LA). Surrounding the shaft portion 241 of the body connector 240, but not overlapping the outer housing 201, is a collar or sleeve 290, also in the shape of a cylindrical tube. The collar 290 is held between the lip portion 242 of the body connector 240 and the outer housing 201 of the body, which together prevent movement of the collar 290 in the axial direction (i.e., parallel to the axis LA). However, the collar 290 is free to rotate about the shaft portion 241 (and thus the axis LA).

[0052] As previously mentioned, cap 225 is provided with an intake hole through which air can flow when a user inhales on mouthpiece 35. However, in some embodiments, the majority of the air that enters the device when a user inhales flows through collar 290 and body connector 240, as shown by the two arrows in FIG.

[0053] 5, an e-cigarette 10 (or more generally any delivery device as described elsewhere herein) may be adapted to operate within a broader delivery ecosystem 1, in which many devices may be in communication with each other, either directly (as indicated by the solid arrows) or indirectly (as indicated by the dashed arrows).

[0054] 5, exemplary delivery devices include an e-cigarette 10 that may be in direct communication (e.g., by using Bluetooth® or WiFi Direct®) with one or more other classes of devices, including, but not limited to, a smartphone 100, a dock 200 (e.g., a home refill and / or charging station), a vending machine 300, or a wearable 400. As discussed above, these devices may cooperate in any suitable configuration to form a delivery system.

[0055] Alternatively or additionally, a delivery device, such as the e-cigarette 10, may be adapted to indirectly communicate with one or more of these classes of devices over a network, such as the Internet 500, for example using WiFi, near field communication, wired links, or integrated mobile data. Again, as discussed above, these devices may cooperate in any suitable configuration to form a delivery system.

[0056] Alternatively or additionally, a delivery device, such as an e-cigarette 10, may indirectly communicate with the server 1000 over a network, such as the Internet 500, for example by using WiFi itself, or through another device in the delivery ecosystem, such as a smartphone 100, a dock 200, a vending machine 300, or a wearable 400, for example by using Bluetooth or WiFi Direct, which then communicates with the server to relay the e-cigarette's communications or report on its communications with the e-cigarette 10. Thus, other devices in the delivery ecosystem, such as a smartphone, a dock, or a point of sale / vending machine, may optionally act as a hub for one or more delivery devices that are only capable of short-range transmission. Such a hub may thus extend the battery life of delivery devices that do not need to continuously maintain a WiFi or mobile data link. It should also be appreciated that different types of data may be transmitted with different priorities. For example, data relating to a user feedback system (such as user factor data or feedback behavior data as discussed herein) may be transmitted with a higher priority than more general usage statistics, and similarly, some user factor data relating to shorter term variables (such as current physiological data) may be transmitted with a higher priority than user factor data relating to longer term variables (such as current weather or day of the week). A non-limiting example of a transmission scheme that allows for high and low priority transmission is LoRaWAN.

[0057] However, other classes of devices in the ecosystem, such as smartphones, docks, vending machines (or any other point of sale system), and / or wearables, may also communicate indirectly with server 1000 over a network such as the Internet 500 to fulfill an aspect of their own functionality or on behalf of the delivery system (e.g., as relay or co-processing units). These devices may also communicate with each other, either directly or indirectly.

[0058] In one embodiment herein, to configure a user feedback system as described below in this specification, the server 1000, the delivery device, such as the e-cigarette 10, and / or any other device in the delivery ecosystem may utilize one or more information sources in the delivery ecosystem or accessible by one or more devices thereof to more accurately respond to the user's condition. These may include information sources such as a wearable or a mobile phone (or any other information source such as a dock or vending machine) or a storage system 1012 of the server. The delivery device may also provide information (such as data regarding interaction with the e-cigarette) to one or more data receivers in the ecosystem, which may again include one or more of a wearable, a mobile phone, a dock or vending machine, or a server.

[0059] To configure a user feedback system as described later in this specification, devices in the delivery ecosystem, such as the delivery device 10, may utilize one or more processors to analyze or process this information to estimate the state of a user (whether of a normal / default user, a user with attributes similar to the current user, or the current user specifically) and / or a form of feedback action determined to change the estimated state of the user, for example by modifying one or more actions of the delivery device or another device in the delivery ecosystem.

[0060] Of course, a delivery ecosystem may comprise multiple delivery devices (10), for example because a user may own multiple devices (e.g., to facilitate switching between different active ingredients or fragrances) or because multiple users may share at least a portion of the same delivery ecosystem (e.g., users living in the same home may share a charging dock while having their own phones or wearables). Optionally, such devices may also communicate directly or indirectly with each other, with devices in the shared delivery ecosystem, and / or with a server. In such cases, a PIN, ID, or account may be associated with each delivery device so that the device can be associated with the correct user, especially when multiple users share the same delivery ecosystem.

[0061] It should be understood that reference to a "user state" includes one of many states of the user, or equivalently, an aspect of the user's overall state. Thus, for example, a user's stress level, as well as a combination of social context and cortisol levels, as a non-limiting example, is an example of a "user state," but does not fully define the user. In other words, a user state is a state that is relevant to the potential intervention of one or more feedback actions, as described elsewhere herein.

[0062] User Feedback System Now referring to FIG. 6, in one embodiment herein, a user feedback system 2 for a user of a delivery device in a delivery ecosystem 1 comprises an acquisition processor 1010 operable to acquire one or more user factors indicative of a user state, an estimation processor 1020 operable to calculate an estimate of the user state based on one or more of the acquired user factors, and a feedback processor 1030 operable to select a feedback action for at least a first device in the delivery ecosystem in response to the estimation of the user state such that a change in the estimated state of the user is expected.

[0063] FIG. 6 illustrates, by way of non-limiting example, one possible embodiment of such a user feedback system.

[0064] In this embodiment, the acquisition processor 1010, the estimation processor 1020, and the feedback processor 1030 are located in the server 1000. However, it will be appreciated that any one or more of these processors may be located elsewhere in the ecosystem 1, or their roles may be shared by two or more processors of the server and / or ecosystem. For example, the acquisition processor may be located in the e-cigarette or the mobile phone, and the feedback processor may be located in the vending machine or the e-cigarette, or the functions of these processors may be shared between the server and such devices. In other examples, these processors may be available only in the delivery device (e.g., the e-cigarette) or only in the delivery system with the delivery device and the mobile phone.

[0065] Retrieval Processor The acquisition processor 1010 acquires or receives one or more user factors contained in one or more data classes from one or more sources.

[0066] Such user factors may have a causal and / or correlative relationship with the user's state, or some other predictable relationship. Such states may be associated with what is colloquially referred to as the user's "mood," although the user's subjective mood, per se, is not the primary consideration of the feedback system. Rather, the feedback system is concerned with the correspondence between the captured user factor(s) and the user state, the user state, and forms of feedback actions that may alter such state of the user in a predetermined manner, usually beneficial to the user.

[0067] Furthermore, it should be appreciated that if there is a correspondence between user factor(s) and state, state, and feedback, then in principle there is also a correspondence between user factor(s) and feedback, without necessarily having to explicitly infer an intermediary state.

[0068] Classes of data acquired by or to the acquisition processor include, but are not limited to, indirect or historical data, neurological or physiological data, contextual data, environmental or deterministic data, and usage-based data.

[0069] Indirect or historical data Indirect or historical data provides background information about the user that is not necessarily related to the immediate situation (eg, not the immediate environment or context) but that may affect the user's state.

[0070] Examples of indirect or historical data include, but are not limited to, a user's purchasing history, previously entered user preference data, or normal behavioral patterns. Thus, more generally, user choices or actions are typically related to the delivery device, but typically do not result directly from use of the delivery device itself.

[0071] Optionally, such information (or indeed any persistent information, such as preferred user settings, user state and / or feedback behavior model data as described elsewhere herein, account details, or other stored user factor data) may be transferred between devices as a given user purchases or uses different delivery devices, so that such information does not have to be reacquired for the new or each device. Such information may be transferred or shared, for example, by direct data transfer over a Bluetooth link between the old and new devices. However, since a potential reason for purchasing a new device is loss of the old device, alternatively or additionally to the above, the information may be transferred or shared by (similarly) keeping the information remotely in association with an account / user ID and subsequently associating the user's different delivery device / system as well. Thus, the system, including indirect or historical data learned / acquired on the old device, may be adapted for transfer or sharing to the new device, either directly between devices or via a centralized user account.

[0072] As an example of historical information, purchase history may be indicative of a user's state, which may be indicative of the user's general state over time (e.g., in terms of significant or recurring purchases) and / or the user's more recent state (e.g., in terms of recent purchases or purchases whose impact on the user may still be considered).

[0073] Thus, a purchase history that may be indicative of a user's state includes the type(s) of product(s) purchased, frequency of purchase, etc. (not necessarily limited to products directly related to the delivery device or its consumables), purchase method (e.g., online vs. in-store), and purchase amount over a period of time or in a single purchase event. The correspondence between purchase methods (and purchased products or services) that affect the user's state can initially be determined on a population basis (e.g., by questionnaire to allow for matching of a statistically significant amount of data), or on a population subset and / or individual user basis that are similar in attributes to the user. For example, purchases can aid in this process by marking them as associated with a particular state, whether using human-readable markings (which the user then enters on their respective phones, e.g.) or machine-readable markings (e.g., QR codes that the user scans on their respective phones, etc.). If a purchase, such as a consumable, includes a machine-readable marking, this can be registered as an indication of mood.

[0074] Similarly, the consumable may be provided with a means for being recognized as indicating a mood when inserted or loaded into a delivery device, such as a microchip containing a code or another uniquely identifiable means for electronically detecting the type of payload (e.g., a binary pattern of conductive dots on the surface of the consumable that can be detected by corresponding contacts on the delivery device). Such identifiable types may differ by composition (e.g., flavor, active ingredient, or concentration of either) or by predefined administration (e.g., two types may be identical except that they present different heating profiles to the device resulting in different inhalation effects).

[0075] The acquisition processor may obtain indirect or historical data from many sources, including, for example, previously entered user preference data and / or similar logs of interaction and / or usage patterns, web or internet based data 110 such as purchase records received from partners such as vendors, user profile data held in the server storage 1012, information collected with consent by the user's mobile phone 100 relating to various aspects of entered user preference data, online purchases, interaction / usage data (e.g. where the phone operates in tandem with a delivery device such as an e-cigarette as a delivery system available only to the user), user surveys, etc. Similarly, alternatively or additionally, the acquisition processor may obtain such data from the delivery device itself.

[0076] Neurological and / or physiological data Neurological and / or physiological data describes the physical state of the user, with respect to mind and / or body. This data may describe the user's state on various time scales, such as immediate status or state changes (e.g., heart rate), long-term status or state changes (e.g., hormone cycles), or chronic status, such as health level.

[0077] Non-limiting examples of long-term data include indicators of a user's metabolism, body type (e.g., lean, average, obese) or body mass index, chronic diseases, pregnancy, and any other long-term conditions, as well as activity / fitness levels, for example on the order of months to years.

[0078] Such data may be obtained by or to the acquisition processor from one or more user surveys (e.g., surveys filled out specifically to aid in the user feedback system and / or surveys filled out for any third party partners (e.g., fitness wearable devices or social media providers)), opted-in medical or insurance records, or at least in part from other devices such as the fitness wearable 400 and / or other devices of the broader ecosystem 1 such as smart scales. The optional timing of such surveys is described elsewhere herein.

[0079] Non-limiting examples of medium to long term data include the user's hormone levels or hormone cycles, such as estrogen, testosterone, dopamine, cortisol, etc., any acute conditions or illnesses, and activity / health levels, for example on the order of weeks to months.

[0080] Non-limiting examples of medium term data include, for example, on the order of days to weeks, the user's sleep cycles, acute conditions or illnesses, and the user's hormone levels or cycles, such as estrogen, testosterone, dopamine, and cortisol.

[0081] Non-limiting examples of medium to short term data include, for example, on the order of hours to days, the user's alertness, activity, appetite or satiety, blood pressure, body temperature, and again, acute conditions or diseases, and / or hormones.

[0082] Such mid-term data (long or short) may also be obtained by or to the acquisition processor from questionnaires, medical or other records, fitness or other smart devices. Thus, for example, hormone levels may be obtained or inferred from questionnaires, medical or other records, consented diary or calendar entries, and / or fitness or other smart devices (e.g., pinprick blood tests, etc.). Similarly, blood pressure, temperature, activity levels, etc. may be obtained from a smart device (usually wearable) or user input.

[0083] Non-limiting examples of short-term data include, for example, on the order of minutes to hours, the user's sweat response, galvanic skin response (phasic and / or tonic), activity level, appetite or satiety, blood pressure, respiratory rate, body temperature, muscle tension, heart rate and / or heart rate variability, and again, any acute conditions or diseases, and / or hormones.

[0084] The acquisition processor may also acquire neurological and / or physiological information specific to the delivery device, such as the cumulative amount of vapor generated over a short period of time (e.g., a period of time equivalent to 1, 2 or more times the pharmacological half-life of the active ingredient in the user's body).

[0085] Non-limiting examples of recent data include, for example, on the order of seconds to minutes, the user's body position, blink rate, respiration rate, heart rate, heart rate variability, brainwave patterns, galvanic skin response (e.g., phasic), muscle tension, skin temperature, voice (e.g., volume, pitch, quality of breath, etc.), and activity.

[0086] Short-term and recent data may also typically be acquired by or to the acquisition processor, for example by biosensing using a smart device or any suitable technique described herein. For example, galvanic skin response may be measured by electrodes on a delivery device, and heart rate may be acquired by optically scanning blood vessels on the wrist with a wearable device or by using an electrocardiogram (ECG) or other dedicated strap-type device. Similarly, brain wave patterns may be detected by an electroencephalogram (EEG), and muscle tension may be detected by an electromyogram (EMG), while body position, eye blinks, etc. may be captured by a camera on a phone or vending machine, for example.

[0087] Short-term and instantaneous physiological data may be difficult to obtain. For example, a smartwatch may only check the user's heart rate every 10 minutes, so for a given puff, the available data may be too old to have direct relevance. Similarly, galvanic skin responses may not be available if the user is wearing gloves. Other data requires the user to select and wear the appropriate sensor, which may not always be the case. Similarly, some short-term and instantaneous data may indicate multiple states of the user, as some values ​​may have multiple causes. For example, an elevated heart rate may indicate a user's stressed state, while another may indicate that the user is enjoying exercise. As a result, such short-term and instantaneous physiological data may only be used optionally in conjunction with other contextual data that helps differentiate readings. For example, an elevated heart rate during work hours, when the user is at work and not traveling significant distances, is much more likely to be an indication of stress than the same heart rate on a Saturday morning, when the user is detected moving at a running pace near his or her home.

[0088] Thus, short-term and instantaneous physiological data (and, if necessary, any neurological or physiological data for which context segmentation may be required) may be ignored or weighted less in the acquisition and packaging of data and / or evaluation of the user's condition, or, optionally, may be used or weighted more fully where segmentation contextual data is also acquired that indicates the type of condition to which the data may be relevant.

[0089] It will be appreciated that in the above description, to the extent that the same example spans different time frames, e.g., different hormones, hormone cycles, fitness levels, etc., may have shorter and longer term characteristics. It will also be appreciated that if an example of data is included in one list but not in another, this does not preclude collection / use of that data over different time frames. For example, blood pressure is given as an example of short term data, but may clearly be part of long term data, e.g., due to persistent hypertension.

[0090] As with indirect or historical data, any suitable combination of data types and / or data from multiple sources may be used.

[0091] In addition to directly measured neurological or physiological data, any suitable analysis or data fusion may be performed to obtain data regarding the user's condition that is specifically relevant to the delivery device.

[0092] For example, the feedback system may be operable to estimate the current nicotine concentration (as one non-limiting example of an active ingredient) or the concentration of an active or inactive compound that breaks down in the user's body from the consumed ingredient (and then delivers the nicotine / active ingredient accordingly).

[0093] Thus, in principle, a feedback system (e.g. in a pre-processor or subsystem of the acquisition processor) may estimate the concentration of nicotine in the user based on monitoring the consumed nicotine, the time of consumption, and the value of the half-life of nicotine in the body (around 2 hours, but this value can be refined based on personal information such as height, weight, etc.). Such monitoring can be performed based on usage data from the delivery device. Thus, for example, based on the original active ingredient concentration and a pre-defined relationship between the heating / aerosol generator power and the aerosol mass output, the mass of active ingredient per unit volume of inhalation may be estimated, from which the amount of absorbed active ingredient may be determined using a pre-defined absorption relationship (optionally based on an analysis of the depth / duration of inhalation with airflow data). Finally, the concentration of active ingredient and / or degradation products in the user over time may be determined by using the user's obesity level and potentially other factors such as age, sex, etc. Again, nicotine is a non-limiting example of an active ingredient.

[0094] It has been found that users typically seek to maintain nicotine levels between upper and lower thresholds (which may vary from user to user), which may collectively define a "baseline" level. The feedback system may establish such a baseline (e.g., by monitoring the user over time) and may select one or more operations of the delivery device to deliver nicotine, as described in more detail below, and optionally modify the operations to match the baseline. The baseline may be a stable value or may vary, for example, by time of day or day of the week. The baseline may be initially estimated based on a profile of the user, e.g., obtained from a questionnaire, and / or may be established or refined with information from the user (measurement and / or self-report).

[0095] Because it has been found that a user's chances of feeling positive are increased when their nicotine levels are closer to an individual's baseline or threshold range, such modifications may be expected to positively alter the user's estimated state.

[0096] If the user consumes multiple different active ingredients, each may have its own baseline threshold. Optionally, the feedback system may monitor whether there is overlap with another active ingredient to the extent that consumption of one active ingredient may affect the baseline of another active ingredient, and if so, may modify these accordingly, for example based on stored pharmacokinetic data associated with such overlap.

[0097] As noted above, in these circumstances a user may interact with multiple delivery devices to consume different active ingredients, with use from each device being combined for the associated user. Alternatively, if a single device is capable of switching payloads (e.g., dispensing different gels) or has a mixed payload of active agents, the currently heated payload or payload mixture can be communicated to a feedback system for purposes of tracking consumption.

[0098] Context Data Contextual data relates to situational factors other than environmental factors (see elsewhere herein) that may affect the state of the user. Typically, such situational factors affect the user's psychological state or disposition toward stress, neutrality, happiness, sadness, or toward a particular behavioral pattern, and therefore may also affect and / or correlate with neurological and physiological user factors such as dopamine or cortisone levels, blood pressure, heart rate, etc., as described elsewhere herein.

[0099] Examples of contextual data include, broadly, place of residence, religion (if any), and, more narrowly, the user's culture, such as work and / or employment status, educational background, and socio-economic factors that may interact with these, such as gender and relationship status.

[0100] Such information may be obtained by or to the capture processor from user surveys, social media data, etc. The optional timing of such surveys is described elsewhere herein.

[0101] Other contexts include seasons (e.g., winter, spring, summer, fall) or months, as well as any particular events or periods within those seasons or months (Lent, Easter, Ramadan, Christmas, etc.) For example, users may be more likely to view below-personal baseline consumption in a positive light during Lent or the first few weeks of January.

[0102] Such information may be obtained by or to the acquisition processor from calendars and databases of events suitably filtered as required according to other contexts such as country, religion, employment, gender, etc. as discussed above.

[0103] Other contexts include a user's agenda or calendar, which may indicate sources of stress or relaxation as well as how busy the user is at a given time. Thus, for example, a social event may be associated with a positive effect on the user's state, such as increased dopamine levels, while a doctor's appointment or driving test may be associated with stressors, such as increased cortisol and heart rate. Similarly, a rapid succession of events, appointments, and / or reminders may have a negative effect on the user's state. The nature of events in a user's agenda or calendar may be determined by keyword analysis, as described elsewhere herein, and frequency or other criteria associated with the calendar, such as the number of appointment conflicts, or individuals likely to attend meetings, may be determined from data provided by the calendar itself.

[0104] The user's schedule or calendar can also indicate where the user is likely to be, which may affect the user's condition or ability to use the delivery device to potentially modify that condition. For example, a user may have different typical conditions and different abilities to use the delivery device depending on whether they are at home, at work, in an outdoor or indoor public space, in an urban or rural environment, or commuting to work.

[0105] The relationship between user states and locations may be based, at least initially, on data from the user's corpus. Alternatively or additionally, the relationship may be built or refined based on data from the user (e.g., measurements or self-reports), so that the user feedback system learns what states the user is likely to be in at a given location, regardless of whether the user has explicitly commented on them.

[0106] Of course, the user's location may be determined from GPS signals acquired by the delivery device or an associated device such as a smartphone, or from the registered location of a vending machine or point of sale unit. Optionally, the device such as the user's mobile phone may include an offline map or location database that records, for example, the location of the user's home and work, and optionally the location of devices that may be temporarily integrated into the user's delivery ecosystem, such as vending machines, public charging units, etc. This allows the device to compare the user's current location against locations that are not considered public (in the case of the user's personal information) or unknown to the user (in the case of additional delivery ecosystem infrastructure). On the other hand, if the user's location is fed back to a backend server for any reason (e.g., as part of data acquisition, combining, packaging, user state estimation, or feedback processing), the user's actual location may optionally be partially obscured (i.e., partially anonymized) by classifying it into a location class such as "work", "home", "gym", "commute", etc., or "other" if unknown. These may also be done on the user's phone, for example, with a predefined tolerance around the classified GPS location associated with the classification.

[0107] Alternatively or additionally, wireless environments may be associated with locations, for example car Bluetooth or train Wi-Fi may be associated with "commute", while home, office or gym Wi-Fi may be associated with respective locations.

[0108] The user may be asked to identify / classify these locations / radio environments by phone, or, optionally, classification may be done automatically by transmitting GPS or radio data to a secure server for comparison, for example, with a database of locations and / or wireless access points maintained by a third party.

[0109] In any case, the acquisition processor may then provide a classification of the location, rather than the location itself, or the acquisition processor may provide a classification of the location.

[0110] For travel modes such as commuting, the type of travel can affect the user's condition. For example, walking can have a more favorable effect on the user's condition than driving in terms of heart rate, blood pressure, etc. More generally, activity levels can affect the user's condition, with increased activity generally having a positive effect on the user, usually during the activity, potentially for the rest of the day, and possibly even the next day. Of course, this context shows that a combination of contexts can be important, as walking in the sun and walking in the rain can have different effects on the user's condition. The type of travel can be inferred, for example, from GPS data on the user's phone, pairing of the phone or delivery device with a vehicle, purchasing a public transport ticket, or a questionnaire indicating travel habits / times.

[0111] Such information may be obtained by or to the acquisition processor, for example from a work or personal digital calendar on the user's phone, and it will be appreciated that the user's phone or other smart wearable may also directly indicate the user's location and / or historical location patterns, for example corresponding to the user's home and work locations and average commute times.

[0112] Other contexts include the weather at the user's location or future weather at the user's location or future locations. For some users, good weather may improve the user's mood and sociability, while bad weather may lower the user's mood and make them less sociable, or affect sociability. For example, some users may behave in a way that consumes active ingredients to an extent that reflects an expectation of the mood suggested by the weather, optionally in conjunction with other contextual and other user factors as described herein.

[0113] Such information may be obtained by or to the acquisition processor from a weather app that may be present on the user's smartphone 100 or that may be directly accessible, for example, by server 1000. More generally, weather data may be obtained in response to GPS data (e.g., by a smartphone) or a location indicated in the user's calendar / schedule and / or using a local weather measurement sensor such as a barometer.

[0114] Other contexts include the user's proximity to others, generally in crowds or social environments, and specifically with respect to other individuals where there is in principle a measurable correlation with the user's behavior. For example, a user may be in a different state depending on whether they are close to their boss, colleagues, friends, partner, children, or parents. Thus, for example, a user may be in a different state in a crowd or social environment than when they are alone or with their partner or family. Or, if children are present, they may behave differently than if there are only adults around.

[0115] Such proximity can be inferred from the user's schedule or calendar, mobile phone, delivery device, or location. A user may self-report their social status generally, for example on social media, specifically for purposes of the user feedback system herein. Alternatively, for example, a phone and / or delivery device may detect signals from other phones and / or delivery devices to indicate that they remain in each other's presence for more than a predefined period of time. Optionally, the other may be detected by use of the phone's camera, which may not be available if the phone is in a pocket or bag. The feedback system may also determine the proximity of a user of a delivery device to other users of such delivery devices (e.g., any suitable delivery device whose location may be determined by the feedback system (e.g., directly or via an associated mobile phone)), whether or not the other delivery devices are part of the feedback system itself. Similarly, the feedback system may determine the proximity of certain people to the feedback system that the user has identified with their permission, for example, by providing the feedback system with a phone number or by associating a detected Bluetooth or other ID with the user.

[0116] Users may also indicate (e.g., via a questionnaire) their typical state in response to various social situations, groups, or individuals, whether at a broad level such as "introvert" or "extrovert," or at a more specific level.

[0117] Other contextual information about the user's social situation can be measured directly using sensors such as microphones or cameras.

[0118] For example, the microphone of the delivery device, the user's mobile phone, or any other device in or connectable to the delivery ecosystem may be used to detect the user's voice (e.g., when specifically speaking to the device or to nearby others, during a phone call, or optionally as an ongoing background activity similar to a voice-activated personal digital assistant). Analysis of the characteristics of the user's voice, such as volume, speech rate, timbre, tone, pitch, and / or inharmonic content, may optionally determine whether the user's vocalizations are normal or stressed, for example after calibration against the user's neutral voice. Similarly, such delivery ecosystem devices may optionally monitor keywords indicative of various states of the user (positive and / or negative).

[0119] Such use of microphones also allows for background noise / music detection. A lot of random or non-music noise may indicate a high stress situation, such as commuting, while a quiet background may indicate a calm situation, such as at home or in the garden. Similarly, the type of music being listened to, whether specifically identified using known techniques or analyzed to determine the presence or absence of beats and tempos, may indicate the state of the user, with certain tracks associated with positive or negative emotions indicating the corresponding emotional state of the user. For example, upbeat dance music may indicate high dopamine levels, while slow songs or songs in minor keys may indicate a depressed mood or depression.

[0120] As with any other functionality described herein, if a suitable sensor (in this case a microphone) already exists that provides the other analysis, this additional mechanism could also be provided by a software update to one or more of the associated devices.

[0121] Similarly, devices in the delivery ecosystem or connectable to the delivery ecosystem may be equipped with cameras from which images can be obtained data regarding the state of the user, such as the user's overall facial expression, which typically correlates strongly with the user's subjective mood, as well as detection of when the user is alone or in a crowd, or with specific individuals, which may have a strong correlation with the user's state, as described herein above.

[0122] Of course, alternatively or additionally, such cameras may be adapted to use other types of data, such as physiological data as described herein above. For example, facial muscle tension may be detectable, which tends to correlate with stress, tension, or pain, while eye movements may indicate the user's concentration and / or the nature of the user's activity (e.g., eye movement and / or blinking patterns tend to be different when driving, reading, or socializing, and when alert versus drowsy). Similarly, micro-movements of the face or neck, if the camera is capable of resolving them, may indicate heart rate.

[0123] Of course, there are other contexts that may affect the user's state, such as recently consumed information (social media content, websites, internet searches, online advertisements, news articles, streaming videos, e-books, e-magazines, photos, music, and other similar content that may be obtained by or to the acquisition processor. Some content may be assumed to have a universal effect on the user's state, such as news of natural disasters, while other content may have a different impact on an individual, such as the performance of a user's favorite sports team, and may be evaluated individually, for example, based on the results of a user survey.

[0124] The content of the consumption information may be such that a rating of its positive or negative impact on the user's condition is generated, for example by a keyword rating. Optionally, only ratings may be obtained by or for the acquisition processor, or any suitable digest of keyword selections etc. More generally, the acquisition processor may only receive a digest of user factors as appropriate, especially if the material itself does not list any user factor characteristics.

[0125] Such evaluation may be performed by the acquisition processor or a processor managing the content as described above, in which case the processor may provide the results directly to the acquisition processor or to a device in the delivery ecosystem where the content is consumed, which may then be relayed back to the acquisition processor, optionally after further processing, for example to incorporate one or more other criteria measured within the delivery ecosystem.

[0126] As a specific example, when a user consumes social media via, for example, a mobile phone, the social media platform, the consuming device such as the mobile phone, and / or the acquisition processor (whether operating separately from the mobile phone or at least partially implemented in the consuming device such as the mobile phone) may optionally analyze the text of posts read by the user, and in particular, optionally, the text of posts written by the user, for keywords. Some keywords may only indirectly indicate the user state. For example, the user's choice of adjectives or verbs in a sentence may indicate whether the user is feeling positive or negative, or neutral or stressed. Similarly, swearing may indicate stress. On the other hand, keywords may explicitly indicate the user state, such as "feeling anxious about going to the dentist." In this case, the word "feeling" indicates that the user is talking about himself, and the keywords "anxious" and "dentist" may be considered as part of a predetermined set of keywords that are usually negatively connoted or that may determine a correlation or correspondence with stress levels. Analysis of social communications consumed or generated by a user may optionally include analysis of one or more of the user's word choice, profanity, or explicit displays, as described above.

[0127] It will also be appreciated that, as an alternative or in addition to the use of social media, the above techniques may optionally be applied to telephone or real-world conversations with others in conjunction with a user's text and / or calendar entries, as well as voice recognition.

[0128] Similar techniques may be applied to metadata associated with news articles, e-books, e-magazines, etc., as well as streaming video, photos, and / or music.

[0129] Thus, for example, while the background noise around the user or the song may be evaluated using the microphone as described above in this specification, in contrast, when the user selects a song, and in particular when listening to a song, song-related metadata may be obtained including one or more of the artist, title, genre, and lyrics, any of which may be based on, for example, a server-maintained database or heuristics similar to those described above with respect to the use of the microphone, and evaluated based on the supplied metadata to indicate a corresponding user state.

[0130] Similar to rating songs, other content such as websites can be similarly rated. Some websites are intended to entertain, while others are intended to inform, and therefore can be categorized according to genres or, more generally, classifications that correlate or correspond to various user states. As described elsewhere herein, website content can likewise be parsed for keywords.

[0131] Similarly, use of devices other than the delivery device may affect the user's state. In particular, the user's choice of apps on the phone, interactions with the apps, types of interactions, and / or duration of interactions with the apps may correlate with the user's state. For example, playing a social media or gaming app may increase dopamine and / or cortisol levels, heart rate, etc., while listening to a music app may change heart rate and / or cortisol levels. The duration of interactions may have a linear or non-linear relationship with these state changes, or may indicate different states over time. For example, playing a game for an extended period of time may indicate boredom.

[0132] Of course, for many user factors (not just context but other types as well), the situational response (e.g., expected state) may, at least initially, be based on data from a cohort of users (e.g., a prior test population of users), but may alternatively or additionally be built or refined by information obtained from the user (whether measured, received, or self-reported).

[0133] Environmental and deterministic data Environmental and deterministic data effectively relate to long-term contextual data outside of the user's preferences or influence, and overlap with longer-term contextual influences such as culture (and thus, for example, the user's upbringing, genetics, gender, internal biome (e.g., gut biome) and / or external biome (e.g., dry or lush living environment), and age).

[0134] As with other data described herein, such environmental and deterministic data may be obtained by or to the acquisition processor from one or more user questionnaires (e.g., questionnaires filled out specifically to aid in the user feedback system and / or questionnaires filled out for any third party partners (e.g., fitness wearable device or social media provider)). Among other things, such questionnaires may ask details such as gender, height, weight, ethnicity, age, etc. Such questionnaires may also include psychological test questions to estimate the user's psychological disposition and / or background (e.g., one or more of extrovert / introvert, active / passive, optimistic / pessimistic, neutral / anxious, independent / dependent, satisfied / depressed, etc.). Such questionnaires may also ask questions relating to the user's culture and beliefs (e.g., one or more of country of origin of the user or their parents, religion (if any), political beliefs (if any), newspapers or news websites subscribed to (if any), consumption of other media (if any), etc.). Again, as with other data described herein, some of such environmental and deterministic data may be obtained by or to a capture processor from medical or insurance records, with consent, and / or may be inferred from the user's location, as appropriate. The optional timing of such surveys is described elsewhere herein.

[0135] Not all environmental and deterministic data need to be long-term, e.g., time of day, day of week, and month are also considered environmental and deterministic data. Thus, for example, user state may vary during the day or week (e.g., weekdays vs. weekends and / or weekday working hours vs. nights), or during certain times of the day. Similarly, it may overlap with other contextual data, e.g., weather. Again, synergies between different user factors are possible. For example, the time of year may affect the amount of sunlight (both in terms of duration and possibly weather patterns). The level and / or duration of sunlight may have a detectable relationship to the user state, although it may be measured (e.g., using a light sensor / camera on the device in the delivery ecosystem) or inferred from the date. Light quality (e.g., color temperature, indoor / outdoor flicker) may also be treated as such a user factor (e.g., captured by a camera in the delivery ecosystem).

[0136] Of course, time of day, day of week, and month are also considered contextual data, especially when functionally defined as work hours or routines, commute time, personal time, quality time, etc., or when patterns of behavior over days, weeks, or months are established. These relate to changes in user factor data as a function of day, week, or month, or the relationship of such user factor data to user state or discriminatory feedback actions. Thus, as described elsewhere herein, one-stage or two-stage models relating user factor data to feedback actions may use time as an input, or different models and different times may be used.

[0137] Usage-based data Usage-based data relates to a user's direct interactions with the delivery device and / or optionally any other device in the delivery ecosystem or devices capable of reporting interactions to a feedback system (e.g., an acquisition processor). These interactions may be related to vaping / consumption and / or device operation / handling and / or settings.

[0138] Vaping / consumption-based interactions may relate to the number, frequency, and / or distribution / pattern of puffs / consumption behaviors within one or more selected time periods. Such time periods may include daily, hourly, or any other time period that may be relevant to the user's condition, as a function of location, pharmacokinetics (e.g., the half-life of one or more delivered active ingredients), and / or any other time period selected to increase the apparent correlation between the number, frequency, and / or distribution / pattern of puffs / consumption and the user's condition. Thus, by way of non-limiting example, the total number of puffs over the life of the device, since refilling of the payload / consumable, over a month / week / day / hour, and / or over a particular session, and over one or more corresponding cumulative hours of use, may be included as metadata along with any puff data used for analysis by the delivery device, a companion device such as a smartphone, and / or a remote server. Such puff data or puff metadata may also be associated with a timestamp so that matching of the puff data can be performed later, for example on the phone or a remote server, and / or associated with a timestamp and / or associative link to other (non-puff) data, as described elsewhere herein.

[0139] Vaping-based interactions may also relate to statistical descriptions of individual vaping actions or cohorts thereof (e.g., including but not limited to, cohorts during one of the selected time periods described above), such as duration, volume, average airflow, airflow profile, active ingredient ratios, heater temperatures, etc.

[0140] Data relating to vaping and vaping behavior (or, more generally, consumption) as described above may be acquired by or to the acquisition processor from the delivery device itself, for example via a Wi-Fi connection to a server 1000, or via communication with a local computing device, such as a companion mobile phone 100 that pairs with the delivery device 10 via, for example, a Bluetooth connection to configure the delivery system.

[0141] The delivery device may include one or more airflow sensors as described herein above to determine the timing and / or manner of vaping by a user, e.g., such features, and raw data regarding vaping / consumption events may be stored in a memory of the delivery device or transmitted to a companion mobile phone. The data may then be used by a processor in the delivery device and / or mobile phone to determine the number, frequency, and / or distribution / pattern of puffs / consumption events within one or more selected time periods, and / or characteristics of one or more vaping / consumption events, such as the duration, amount, average airflow, airflow profile, average component ratios, heater temperature values, etc.

[0142] Optionally, at least one sensor may be configured to detect at least two of the puff profile, puff frequency, puff duration, number of puffs, session length, and peak puff pressure, and to determine the user's mood from the detected information.

[0143] A user's puffing behavior provides a useful indicator of stressed versus non-stressed states. In particular, it has been found that the intensity and frequency of a user's puffs change from normal levels (e.g., stronger and shorter puffs, more frequent puffs) during moments of stress. Thus, the delivery device, another device in the delivery ecosystem (such as the user's mobile phone), or a back-end server may optionally establish a baseline profile, frequency, pattern, number of puffs, peak pressure, etc., as described elsewhere herein, and then detect deviations from this baseline (e.g., above a pre-defined threshold) that are indicative of stress. Furthermore, different baselines can be optionally set for different situations and contexts, such as time of day, day of the week, and location. For example, a user may be more stressed at work than at home, but this partial stress increase is considered a baseline for the work context, and only if the user becomes more stressed is a deviation indicative of a change in the user's stress that may prompt a reaction.

[0144] For example, a puff profile characterizes the variation of the draw strength over the duration of the draw. Thus, for example, the airflow rate of the puff may be used to characterize the puff profile, with a larger airflow rate associated with a short and intense draw indicating a higher stress level than a smaller airflow rate. Puff frequency is also considered to be correlated with stress, such that in stressful situations the puff frequency is higher than in normal user situations. Puff duration is considered to be a subset of the puff profile, and to a first approximation, the duration also indicates the type of draw, typically shorter puffs in stressful situations and longer puffs in normal user situations. Peak puff pressure is also considered to be a subset of the puff profile, indicating the intensity of the user's draw.

[0145] The number of puffs within a session may also indicate the state of the user. A session may be understood to be of fixed duration or may be functionally defined as a period of time that includes inhalations separated by less than a predefined period of time required to indicate that the session is over. In any given session, all other conditions being equal, the number of puffs taken by a user may be greater if the user is in a stressed state than normal. Similarly, a session, if functionally defined, may be shorter if the user is in a stressed state than normal.

[0146] In any event, such information may then be packaged as one or more user factors and sent to a capture processor, as described above.

[0147] Manipulation / handling-based interaction may relate to the manner in which a user interacts with the delivery device when not actively vaping, characterizing, for example, whether the delivery device is stored in a bag until immediately prior to use, or whether the user plays with or tampers with the delivery device during use.

[0148] In particular, micro-movements or vibrations between 3 and 9 Hz may be considered to indicate elevated stress or arousal, and detection of such movements may be used as an input indicative of stress. Conversely, for example, low frequency manipulations may optionally be considered to indicate low stress. Similarly, movements that are not "micro" movements (i.e., intentional shaking of the device as opposed to holding it in a trembling hand) may be ignored for the above purposes or may be analyzed in other ways (e.g., movements characteristic of imminent use or associated with a broader context in which a user state has a clear correlation).

[0149] For example, apart from micromotions, a given user may tend to hold a delivery device more frequently and / or for longer in stressful situations than in relaxed or neutral scenarios because the user makes a conscious or unconscious association between the device and the stress relief that vaping with it provides. Such holding may be detected using one or more accelerometers and / or touch sensors, as described elsewhere herein.

[0150] Optionally, the delivery device, another device in the delivery ecosystem (such as the user's mobile phone), or a back-end server may establish a baseline level of holding and / or physical interaction of the delivery device to detect when increased holding and / or physical interaction (e.g., above a predefined threshold) occurs, and may also optionally correlate such normal or elevated levels with other indicators or correlations of the user's state and / or directly with the user's state to allow holding and / or physical interaction to act as a proxy for (another) designation of the user's state and / or another cause of such designation that is normally used / preferred but currently unavailable. As with puff behavior, further optional, different baselines can be set for different situations and contexts, such as time of day, day of week, and location. For example, a user may be more stressed at work than at home, but this partial stress increase is considered a baseline for the work context, and only if the user becomes more stressed is it considered an deviation indicative of a change in the user's stress that may prompt a reaction.

[0151] The delivery device may include one or more touch or motion sensors (e.g., accelerometers) that determine such interactions. Similarly, the device may include buttons or other environmental features that allow for recording of user interactions. Interactions with buttons or other environmental features on the delivery device on a companion mobile phone may also be recorded. Such interaction data may then be packaged as one or more user factors and sent to a capture processor.

[0152] Thus, for example, by using telemetry from one or more motion sensors in such a delivery device, a user feedback system can detect incidental or hidden manipulation of the device based on characteristic orientation changes such as spinning, flipping, rocking, etc. that are not associated with any overall movement of either the device or the user. Such tampering may indicate a state of the user, for example, at least an unconscious desire to use the device or a desire to use the device more than currently, and thus correlate with increased stress, lack of concentration, and / or deviations from a preferred baseline amount of active ingredient in the user's body.

[0153] Similarly, in delivery devices that use a UI interface, such as a button press for activation, the delivery device may measure the time between such activation and the inhalation that occurs. This time may correlate or correspond to one or more of user stress, user fatigue, user concentration, and deviations from a suitable baseline amount of the active ingredient in the user's body. Thus, for example, this time may be shorter when the user is stressed than when they are normal.

[0154] The delivery device or other devices in the delivery ecosystem may be equipped with other sensors, such as microphones or cameras, operable to capture measurements of behavioral usage, which may record inhalation and / or exhalation, the relative timing and characteristic behaviors associated with inhalation actions, or other behaviors and usage, which may be determined by audio and / or image analysis as input to a capture processor.

[0155] Of course, if a user has multiple delivery devices 10, usage may be aggregated across these devices by obtaining user factor dates from each device, or usage may already be aggregated through an intermediary, such as a phone app or one of the delivery devices acting as a hub for this purpose. Also, as a non-limiting example of pharmacokinetics, if different devices deliver different active ingredients (whether by type or concentration), this may be taken into account in modeling usage.

[0156] Multiple Data Sources As discussed above and shown in FIG. 6, the acquisition processor may receive a number of user factors of the type described herein from one or more data sources, such as data sources in the delivery ecosystem 1, data sources in the Internet 110, and records maintained by the feedback system 1012, such as on the server 1000.

[0157] As discussed above, these user factors may be variously categorized as indirect or historical data, neurological or physiological data, contextual data, environmental or deterministic data, and / or usage-based data.

[0158] In the case of usage-based data, it will be appreciated that some or all of such usage-based data may be obtained through the use of multiple sensors and / or sensors having multiple sensing capabilities in a sensor platform.

[0159] Although the above description describes the sensor platform in relation to providing a feedback system, it should be understood that a delivery device, such as an aerosol delivery device, may in any case include such a sensor platform independent of any feedback system, for example to provide information to a user or other device. Thus, any suitable aerosol delivery device may have a sensor platform including one or more selected from the list consisting of a galvanic skin response detector, a heart rate detector, a touch detector, and any other suitable detector described herein (e.g., a cortisol detector), and the or each detector may be located on one or more selected from the list consisting of a grip portion of the delivery device, a mouthpiece of the delivery device, and an activation button of the delivery device, as appropriate. Another location of the sensor platform for the one or more sensors is, for example, on an optional collar attachment configured to fit between the cartomizer and the body of the delivery device, between the body of the device and the mouthpiece, or between any two user-separable components of the delivery device, as appropriate. The collar may be user-attachable, user-removable, or integral to the device. The collar may, for example, be configured to surround a portion of the airflow path in the delivery device in this manner and may include the airflow sensor (or another airflow sensor) for detecting, for example, puff strength or profile, one or more accelerometers or gyroscopes for detecting one or more of the device's orientation, acceleration, speed, or position, or any other suitable sensor (e.g., touch sensor). The collar may, for example, draw power from the delivery device while powering the cartomizer. The collar may also include a Bluetooth link and an antenna (e.g., an annular antenna). In this manner, the collar allows for the retrofitting of functionality related to the technology herein to delivery devices that are otherwise unsuitable or may only provide limited information.

[0160] Of course, different data sources relate to data spanning different time periods. For example, a single puff may take 1-3 seconds, while the work context applies for several hours before and after this event. Thus, the acquisition processor may by its nature use data sources that precede, coincide, and / or follow the occurrence of other data sources, and correlation with the user's state at a particular time (e.g., as indicated by something more transient, such as a puff) may take these broader perspectives into account through any significant correlation.

[0161] However, in addition to sources of data that are inherently long-lasting, it will be appreciated that the techniques described herein can use one or more sensors to obtain data that complements readings from other sensors by recording data before, during, and / or after readings by the other sensors. For example, a heart rate monitor may measure the heart rate before, during, and / or after a puff or series of puffs. Thus, data regarding the user's condition is effectively obtained not only during the actual inhalation (or other measurable interaction) with the device, but also from the periods immediately prior to and after the inhalation.

[0162] Of course, recording of any measurable parameter of interest can provide a time-localized context for the measurement of any other parameter, although it may be beneficial to measure such parameters, particularly on one or both sides of the puffing motion.

[0163] Data from such a sensor or sensors could be analyzed continuously, but this would require high processing resources and power consumption, adversely affecting the battery life of the delivery device. As a result, the delivery device optionally includes a circular memory buffer that is updated with values ​​from the or each sensor used in this manner, thereby minimizing processing. The buffer includes storage for, for example, 5, 10, 15, 30, or 60 seconds of sensor readings, and is overwritten in a circular fashion during use.

[0164] If a puff is detected, buffered data is collected either directly via data transmission or via a companion device for processing on the delivery device or via data transmission to a companion device such as a mobile phone or a remote server, as described elsewhere herein.

[0165] If a puff is detected, then optionally the buffer data is collected for another period approximately equal to half the circular memory duration of the buffer, such that the puff event corresponds to the middle of the resulting buffer record, and data preceding and following the puff event are also recorded. Of course, the balance of preceding and following data recorded in the buffer can be controlled by using the amount of buffer data recorded upon and after detection of a puff event. This can be used depending on whether the sensor data being recorded is particularly relevant to preceding a puff, following a puff, or a balance of the two.

[0166] The recorded data may be relative to the start of the detected puff or the end of the detected puff. Thus, for example, in the case of a 2 second puff and a 5 second buffer, sensor data from one or more sensors other than the airflow sensor may be sent for analysis over the preceding 5 seconds, the preceding 4 seconds and 1 second of puff, the preceding 3 seconds and 2 seconds of puff, the preceding 2 seconds, 2 seconds of puff and the following 1 second, the preceding 1 second, 2 seconds of puff and the following 2 seconds, 2 seconds of puff and the following 3 seconds, 1 second of puff and the following 4 seconds, or the following 5 seconds. Of course, sub-second divisions or greater divisions are possible depending on the sampling rate of the sensors used in the buffer.

[0167] Thus, more generally, the delivery device may be equipped with a circular buffer that periodically records sensor data from one or more sensors other than the airflow / puff sensor, and when the occurrence of a puff is detected, sensor data preceding, during, and / or following the puff event can be provided for analysis, thereby providing time-localized, contextual sensor data in a power-efficient manner.

[0168] It should also be appreciated that, more generally, any device in the delivery ecosystem may include a circular buffer that periodically records sensor data from one or more sensors other than the airflow / puff sensor, and, if a puff is detected, can provide sensor data preceding, during, and / or following the puff event for analysis, where the puff event may be communicated to the or each other device directly, for example via Bluetooth, or indirectly via a companion device such as a phone, which may optionally collate sensor data from one or more devices, potentially including itself, for analysis by the delivery device, the phone itself, another device in the delivery ecosystem, or a remote server.

[0169] How the Capture Processor Works Referring again to FIG. 6, the acquisition processor 1010 is typically part of the remote server 1000 and may receive user factors from different data sources such as the server's own storage / database 1012, online data sources 110, and devices in the user's delivery ecosystem 1 such as the delivery device 10 itself, a mobile phone 100, a fitness wearable 400, a docking unit 200, a vending machine 300, and any other suitable device that may provide information related to the user's condition (such as a voice-activated home assistant, a smart thermostat, a smart doorbell, or other Internet of Things (IoT) device).

[0170] The acquisition processor 1010 may comprise one or more physical and / or virtual processors, and may be located in a remote server and / or functionality may be distributed or otherwise distributed across multiple devices, including but not limited to the user's mobile phone 100, the docking unit 200, the vending machine 300, and the delivery device 10 itself. The acquisition processor may also comprise one or more communication inputs, e.g., via a network connection and / or a local connection to local storage. The acquisition processor may also comprise one or more communication outputs, e.g., via a network connection and / or a local connection, e.g., to the estimation processor 1020.

[0171] The acquisition processor may include a pre-processor or sub-processor (not shown) configured to parse and / or convert acquired information into user factors, which may not be immediately usable as described above. Examples include keyword or sentiment analysis of consumed media to determine a net positive or negative impact on an aspect of the user's state as a user factor, or similarly keyword analysis of a user's calendar to determine locations and events to determine a net positive or negative impact on an aspect of the user's state as a user factor. Other inputs, such as air temperature or probability of rain, may similarly be converted to a scale appropriate for the user factor, e.g. normalized or classified according to their impact on the user's state.

[0172] In this manner, the acquisition processor may be operable to generate and / or relay user factors for input to the estimation processor at various levels of abstraction from the original material.

[0173] Thus, the original data can optionally be enumerated, coded, sorted, formatted, or otherwise processed, or simply passed through, to provide as input to the inference processor, with potentially as many or more inputs as there are original data sources. As will be appreciated from the above discussion, this can result in a large number of inputs.

[0174] Thus, optionally, one, some, or all of the original data may simply be passed through to any evaluation, coding, classification, formatting, or other processing, or even an intermediate user factor generation stage, as an option, of the acquisition processor, if necessary, so that from the transmitted input, a positive or negative effect on a particular subset of user factors that are related to user state but not directly or easily measurable, such as effects on dopamine and / or cortisol, heart rate, satiety, etc., may be determined.

[0175] Similarly, the intermediate user factor generation stage of such an acquisition processor may combine inputs from similar classes to generate class-level user factors for one or more of the data classes described herein.

[0176] Thus, by way of non-limiting examples, indirect or historical data may be aggregated at a given scale as the manner in which users actively modify or update their respective devices or accept such modifications. Neurological or physiological data may be aggregated at a given scale and / or trajectory on that scale as the apparent stress of the user. Contextual data may be aggregated at a given scale as the socially desirable use of the current delivery device. Environmental or deterministic data may also be aggregated by the likelihood that the user will want to use the delivery device in a given time frame, and usage-based data may be aggregated as the frequency or depth of the user's recent use of the delivery device.

[0177] Of course, in practice, raw data from only a portion or one of the classes may be available, and even if data from one class is available, class-level user factors as in the above example may not be generated, or different types of class-level user factors (e.g. different subsets of individual user factors) may be generated depending on the type of data received within that class. Similarly, class-level user factors may be generated as input to an estimation processor in parallel with the individual user factors.

[0178] Contribution values ​​and / or influences from different individual, subset, and / or class level user factors may then be presented as inputs to the estimation processor, with the selection of class, subset, and / or individual user factors being chosen to provide good discrimination between different user states.

[0179] For example, galvanic skin response can provide a good indicator of the user's condition and responds to nicotine as an active ingredient by suppressing the response. This can therefore be an optional candidate for an individual data source to be used as an input to the estimation processor. Other physiological measures that provide good discrimination include muscle tension (EMG), heart rate, skin temperature, electroencephalography (EEG), and respiratory rate. Any of these available can be considered for inclusion as individual data sources, optionally after any evaluation, coding, classification, formatting, or other processing, in the alternative or addition to the above, in any combination with these user factors or other user factors described elsewhere in this specification.

[0180] Similarly, location, social environment, time of day, and hormone levels are all good indicators of a user's state and may be candidates for use as individual sources of data that serve as input to the inference processor.

[0181] Thus, more generally, the user factors may be acquired by or to the acquisition processor, for example as individual, subset, and / or class level user factors, and after any suitable parsing or processing, provided to the estimation processor as individual and / or subset or class values ​​combined with one or more other user factors (e.g., based on weighted contributions, statistical functions, trained machine learning outputs, look-up tables of pre-computed correspondences between values ​​of the acquired data and values ​​of the target user factors, etc.).

[0182] Estimation Processor The estimation processor 1020 is operable to calculate an estimate of the user state based on one or more of the inputs received from the acquisition processor including the acquired user factors or the acquired user factors. The calculation of the estimate of the user state can be an explicit calculation for generating an output reflecting the user's state prior to the generation of a suggested feedback action (considered a two-step process) or an implicit calculation for generating a suggested feedback action expected to change the user's state (considered a one-step process).

[0183] Like the acquisition processor, the estimation processor may comprise one or more physical and / or virtual processors and may be located in a remote server and / or have functionality distributed or otherwise distributed across multiple devices, including, but not limited to, the user's mobile phone 100, the docking unit 200, the vending machine 300, and the delivery device 10 itself. The estimation processor may comprise one or more communication inputs, for example, for receiving data from the acquisition processor 1010. The estimation processor may also comprise one or more communication outputs, for example, for providing suggested feedback actions to the feedback processor 1030.

[0184] Explicit State Estimation In one embodiment herein, the inference processor first explicitly infers the user's state in a first stage in a two-stage process, and then generates suggested feedback actions in response to the inferred state in a second stage. The inferred state itself may be in the form of a single value or category, or may be a multivariate description of the user's state.

[0185] As a non-limiting example of a single-valued state, the estimated state may be: i. the user's stress level; and ii. the degree of benefit that a user is expected to subjectively experience in response to consuming a unit of the proposed active ingredient; and iii. a social flexibility score indicating the ease with which the user's current use of the delivery device would enable them to change their respective states through delivery modifications; and It may also represent:

[0186] As non-limiting examples of state categories, the probable state may be:

[0187] i. One, all, some, or none of a number of state categories may correspond to what is colloquially referred to as mood (e.g., happy, sad, low cortisol, mid cortisol, high cortisol, neutral, stressed, embracing change (e.g., willingness to alter state using the respective delivery device), or rejecting change).

[0188] ii. One of the state classifications is selected to have a clear correlation with either the input from the acquisition processor and / or the available feedback actions, and these classifications do not necessarily fall into assumed categories such as "happiness" or "high cortisol", but rather have classification boundaries that are driven at least in part by responses to the available input from the acquisition processor or the output to the feedback processor.

[0189] As a non-limiting example of a multivariate description of a user's state, the estimated state may include: i. a user's stress level as a function of physiological indicators as well as contextual indicators, and an indicator of current social flexibility based on time of day, location, and / or proximity to specific individuals; ii. Indicators of the user's physiological state based on galvanic skin response and heart rate, as well as current position in their hormonal cycle, and indicators of mental state derived from questionnaires and / or social media analysis; may also include

[0190] Using these examples, the operation of the estimation processor can be illustrated in a non-limiting manner as follows.

[0191] The estimation processor may transform the input data from the acquisition processor into an estimated state through the use of predefined rules, algorithms, and / or heuristics.

[0192] For example, a single-valued state, such as a user's stress level, may be derived by application of a predefined combination of multiple user factors, such as a weighted sum, with the result normalized according to the number of currently available inputs contributing to the weighted sum.

[0193] Similarly, a single-valued state, such as a degree of expected benefit for a user, may be derived by estimating a positive or negative emotional state of the user based on the sum of index values ​​for positive or negative keywords or emotions in recently consumed or created online media and positive or negative values ​​associated with the user's location classification.

[0194] Similarly, the probable condition category may be selected by template matching of the user factor values ​​against pre-defined values ​​indicative of a given category, or similarly by identifying the minimum mean squared error between the user factors and a template of user factor values ​​for each candidate category, optionally with different linear or non-linear weighting of different categories and large errors to reflect their relative importance in the category discrimination.

[0195] Finally, as an example, a multivariate condition may include deriving individual indicators of the condition according to any of the above examples. Thus, as described above, a single value stress level can be generated for each physiological and contextual indicator, and a social flexibility value can be determined based on scores pre-associated with different times, locations, and particular classes of individuals (e.g., partner vs. children). Alternatively, social flexibility classification may be based on template matching against such scores and / or values ​​for the underlying input data.

[0196] Thus, in one embodiment herein, it is also possible to detect whether a user is stressed or highly aroused by evaluating multiple criteria. For example, N criteria are possible (e.g., N=5, but alternatively N=2, 3, 4, 6, or any suitable number). If more than a threshold number of these criteria are met, the user is determined to be stressed or highly aroused. Optionally, there may be multiple thresholds for different degrees of likelihood that the user is stressed or highly aroused, which may affect the selection or extent of subsequent feedback actions. Optionally, a machine learning system may be employed to model such thresholds for different combinations of criteria (e.g., learning different thresholds for different combinations of criteria, and optionally the extent to which each is met, to indicate the user is stressed or highly aroused or the likelihood of being stressed or highly aroused). Such a machine learning system may be provided pre-trained on a normal user (or a trained user profile that best matches the user), and then in a continuous adjustment / ongoing learning mode to update the thresholds based on various user feedback (as described elsewhere herein).

[0197] Exemplary criteria may relate to one or more data sources (e.g., indirect, historical, neurological, physiological, contextual, environmental, deterministic, and usage-based). For example, the criteria may be: the user is at work, the user's heart rate is elevated, the user is exhibiting tremors, the user is holding or handling the delivery device in a characteristic manner, it is a work day, the user is puffing more frequently than usual (e.g., by a threshold amount), the user is puffing harder than usual (e.g., by a threshold amount), the user is typing with stress-related keywords, recent diary entries include high-stress events, bad weather, or the user is in a location that would normally have been left by the current time (e.g., work or late). Of course, the criteria may be a subset of the above and / or other criteria, and some criteria may not be taken into account if, for example, the user does not have a heart monitor, does not agree with the text, diary evaluation, or the user's handling of a particular delivery device is not detected, etc.

[0198] In any event, if M of N such criteria are met, the user may be determined to be in a stressed or highly aroused state. This inference is then passed to a feedback processor. Optionally, the met criteria or a representation thereof may also be passed to the feedback processor so that a feedback action is selected in response to a particular subset of the met criteria.

[0199] Alternatively or additionally, the criteria can be structured as a decision tree, where a set of criteria along the branches of the tree, if met, indicates that the user is presumed to be in a stressed or highly aroused state. In this case, the same criteria may appear along different paths in the tree, but each path typically represents a unique set of criteria that are conditionally met. The paths may be of various lengths, terminating with different degrees of certainty that the user is in a stressed or highly aroused state.

[0200] Alternatively or additionally, the estimation processor may convert input data from the acquisition processor into the estimated state through the use of a look-up table.

[0201] In one example, these lookup tables may simply provide a pre-computed implementation of the above predefined rules, algorithms, and / or heuristics to avoid repeating these calculations on a server or device in the delivery ecosystem that has limited processing power but can act or share the role of an estimation processor, such as the delivery device 10, dock 200, vending machine 300, wearable device 400, or associated phone 100.

[0202] In another example, such lookup tables may provide associations between input values ​​from the acquisition processor and output values ​​of user states, state classifications, and / or multivariate states pre-derived according to any suitable mechanism, such as feedback from extensive user testing, or as described below in this specification, the output of a machine learning system. Again in the latter case, the lookup tables may potentially provide simple replication of the operations of such machine learning systems by recording input and output pairs to common values ​​that can be easily implemented in devices in the delivery ecosystem with relatively low computing power.

[0203] Alternatively or additionally, the estimation processor may model correlations between the input data and the estimated state of the user. Such correlations may be due to causal relationships between the user factors and the user state, or due to tendencies that the user factors have to associate with causes of the user state, typically acting as proxies with a certain probability. Similarly, such correlations may be due to user factors and user states that are both responsive to separate causes or circumstances such that the correlations can be constructed with sufficient repetition. Similarly, such correlations may be due to the user state giving rise to the user factor. Thus, more generally, correlations relate to measurably predictable correspondences between one or more user factors (whether individual, subset, or class-level user factors as output by the acquisition processor) and the user state (whether single-valued, categorical, or multivariate), typically due to causal relationships (in either direction between the user factors and the state), common causes that result in the responses of the user factors and the user states that have at least a reproducible relationship at a statistical level, and / or measurable correspondences regardless of the perception of direct or indirect causal relationships.

[0204] When modeling correlations, the estimation processor can be trained using a dataset that takes as input data corresponding to the above-mentioned outputs of the acquisition processor and includes as target outputs a descriptor of the user's state (whether single-valued, categorical, or multivariate, e.g., based on direct measurements of the user's state and / or self-reports of the user's state).

[0205] Specific means by which such correlations may be derived include any suitable technique for estimating such correlations, such as a correlation map between inputs and outputs, where the link between a particular input and output is strengthened (e.g., by incrementing a connection weight) by simultaneous (or if a time factor is included, within a predefined time window) presentation of the input and output. When trained on the dataset, a new input will activate one or more candidate states that are correlated with the input to a greater or lesser extent, via the connection weights. The most strongly activated candidate state may then be selected as the user state, and such states may be ranked by activation strength. Of course, such a system may provide multiple input values ​​simultaneously, corresponding to subsets of individual and class-level user factors, as described elsewhere herein, and the generated output may correspond to a single-valued state, a classification, or a multivariate state, where many values ​​represent different aspects of the output user state, as described elsewhere herein.

[0206] An example of a correlation map is a neural network, and may take any suitable form.

[0207] More generally, any suitable machine learning system capable of determining correlations or other predictable correspondences between one or more inputs and one or more outputs is contemplated.

[0208] Given the above dataset, such machine learning systems are typically supervised, e.g., supervised classification learning algorithms, e.g., if the user state is a classification, or supervised regression learning algorithms, e.g., if the user state is single-valued or multivariate. Other forms of machine learning, such as reinforcement or adversarial learning, or semi-supervised learning, are also suitable. Furthermore, multiple independent machine learning systems, separately trained on differences or partially overlapping individual, subset, or class-level outputs of acquisition processors, can improve the modeling results by ensembling to accommodate different configurations of raw data, e.g., due to different users' different patterns of ownership of delivery ecosystem devices and different permissions and habits that affect the availability of online information sources. It should also be appreciated that a mixture of different machine learning systems can be used in parallel to generate, e.g., a multivariate state of a user, e.g., one or more different elements of the multivariate description are generated by each different machine learning system. Each of these machine learning systems can be implemented on separate hardware (e.g., based on dedicated neural processors), but more commonly would be implemented on the same hardware (software-based machine learning systems that are loaded and executed as needed).

[0209] However, unsupervised learning algorithms are also contemplated, so that for example associative learning can determine the probability that a user will be in a given state in the presence of a certain input or input pattern.

[0210] Examples of such machine learning systems will be known to those skilled in the art in the form of algorithms and / or neural networks (e.g., logistic regression, neural networks, support vector machines, etc.). However, machine learning may optionally be used to prepare (e.g., pre-process) data in either the estimation processor and / or the acquisition processor. Thus, clustering (e.g., k-means clustering) may be used to classify a diverse set of inputs into class-level user factors of the type described herein above. Such techniques may be used to derive classifications for user states according to the second example of state category classification described herein above, for example in response to inputs from the acquisition processor or available feedback operations of the feedback processor.

[0211] Similarly, a preliminary step in the estimation processor and / or acquisition processor may employ dimensionality reduction, such as principal component analysis, to reduce the number of inputs while retaining information that significantly corresponds to the user state.

[0212] Of course, such machine learning systems can optionally be used to model user states, derive correlations / correspondences between inputs / criteria, and / or determine univariate or multivariate thresholds (e.g., optionally in the case of decision trees) as needed.

[0213] In summary, when the estimation processor generates an explicit estimate of the user state, it uses a repository of correspondence between available inputs from the acquisition processor and the estimated state, which may be embodied in algorithms, rules or heuristics, one or more lookup tables, and / or one or more trained machine learning systems.

[0214] The result in each case is an estimate of the user state, which may be in the form of a single-value, categorical, or multivariate description / representation of the user state, as described herein above.

[0215] On the other hand, the operation of the estimation processor when generating implicit estimates of the user state is described later in this specification.

[0216] Feedback suggestions from estimated states As described herein above, the estimation processor may be adapted to operate in a two-stage process, in a first stage, by the acquisition processor, to estimate a user state from inputs including one or more user factors or data derived from such user factors, and in a second stage, as described below, to generate suggested feedback actions expected to modify the user's state.

[0217] In principle, the second stage may be realized by the feedback processor rather than the estimation processor, or may be shared between the feedback processor and the estimation processor. Alternatively, the feedback processor may simply receive the suggested feedback actions. In any case, the feedback processor may then select a feedback action (defaulting in the case of only one suggestion, or selecting one or more in the case of multiple suggestions), and optionally act to appropriately generate the feedback action or actions suggested by the estimation processor within the delivery ecosystem.

[0218] For purposes of explanation, the second stage is described herein as occurring in the estimation processor.

[0219] The second stage may be chosen for practical reasons, for example, a training set used in modeling the correspondence / correlation between the user factors or their respective derived values ​​by the acquisition processor and the user state may be easier to generate or obtain than a training set used in directly modeling the correspondence / correlation between the user factor based inputs and the suggestion feedback actions, since the user state may be directly measurable or easier to report by the user.

[0220] Similarly, it may be easier to generate training sets that determine correspondences / correlations between measurable and / or self-reported user states and proposed feedback actions based on, for example, user questionnaires ranking feedback actions for a given state and / or subsequent effectiveness of implemented feedback actions in changing the user state toward a more desirable state, such as measured and / or reported by the user. Optional timing of such questionnaires is described elsewhere herein. Typically, a more desirable state is one that improves the user's subjective sense of well-being and / or brings physiological or neurological indicators of the user's state closer to a preferred baseline (e.g., increased heart rate, galvanic skin response, increased skin temperature, and / or reduced respiration rate, etc.).

[0221] The input of the second stage will typically be an estimate of the user's state, represented by a single value, a category, or a multivariate description as described herein above, or a plurality of these if multiple states are estimated (e.g., different activity / correlation strengths in response to the input of the first stage). Optionally, the input to the second stage may also include one or more user factors and / or inputs as provided by the acquisition processor. For example, as described elsewhere herein, certain physiological measurements may be useful indicators / surrogates of the user's state, such as galvanic skin response, heart rate, respiration rate, skin temperature, etc. Thus, one or more of these inputs, or any other input to the first stage, may also optionally be provided to the second stage in conjunction with the or each estimated state.

[0222] In any event, similar to estimating a user state, generating suggested feedback actions may use any suitable mechanism that embodies a correspondence / correlation between an estimated user state and a suggested feedback action.

[0223] As mentioned above, this may include predefined rules, algorithms, and / or heuristics that convert the estimated states into suggested feedback actions.

[0224] For example, a single valued state (such as the degree of stress) may drive a corresponding suggested feedback action, such as increasing the percentage of active ingredient in the unit volume of inhalation of the generated aerosol, which may be achieved by modifying heater, airflow, reservoir, and / or other payload storage settings, etc., and may be managed by a feedback processor, as described later in this specification. The relationship between the degree of stress and the change in active ingredient may be linear or non-linear, or may qualitatively change at different values, such as no change at all at low levels of stress, a linear relationship at medium levels of stress, and an asymptotic relationship at high levels of stress up to a maximum percentage of active ingredient. At or near this maximum, the behavior of the user interface of the delivery device or other devices in the ecosystem may also be modified, such as issuing a warning or calming message to the user's mobile phone.

[0225] Alternatively, for example, a single category state may have a corresponding suggested feedback action.

[0226] Finally, in some cases, a multivariate state may result in a corresponding suggested feedback action based on, for example, weighted or non-weighted contributions from different elements of the state description, and / or different feedback actions may be suggested based on overlapping or non-overlapping subsets of elements of the state description. Thus, for example, if the state description suggests that the user is stressed and is in a work environment, the feedback action may assume that the user is in a work environment and therefore experiencing implicit stress, but is currently unable to increase the intake of the active ingredient, and issue a message to the UI of the delivery device or other devices in the ecosystem, such as the user's phone, to suggest that the user take a break. On the other hand, if the user is stressed but is not in a work environment, the feedback action may be similar to the example stress level described above, resulting in an increase in the rate of active ingredient delivered to the user.

[0227] As mentioned above, any one of these may be accompanied by one or more inputs to the first stage.

[0228] Thus, in one embodiment of the present specification, it is also possible to detect whether a user is in a state of stress or high arousal by evaluating multiple criteria, as described above in this specification in connection with the first stage of the estimation processor.

[0229] In this case, the single-valued state may be an inference that the user is in a stressed or highly aroused state, for example because a threshold number of criteria are met, or an inference that the user may be in a stressed or highly aroused state, for example because many criteria are met.

[0230] A multivariate state, on the other hand, would be or include some or all of the criteria met or data indicative thereof. Thus, for example, the criteria met might be being at work, being later than usual, and having an elevated heart rate. This combination may be determined from a set of met criteria or a sequence of met criteria in a decision tree, as described elsewhere herein.

[0231] This combination of criteria may correspond to a particular feedback behavior, which may be related to non-vaping behavior if vaping is not permitted at the workplace. Different combinations of met criteria may correspond to different feedback actions. As described below, this correspondence may be embodied in rules, lookup tables, or machine learning.

[0232] Again, as with estimating the user state, the estimation processor may alternatively or additionally convert the state estimation data into suggested feedback actions through the use of a look-up table.

[0233] Alternatively or additionally, similar to estimating the user state, the estimation processor may model the correlation between the estimated user state and the suggested feedback actions and use similar techniques to do so.

[0234] If the inference processor models correspondence / correlation, it can be trained using a dataset that includes as input data corresponding to an inferred user state (e.g., in the form of a single value, a classification, or a multivariate description, or a combination thereof), optionally an input from an acquisition processor as described above in this specification, and a suggested feedback action as a target output.

[0235] Suggested feedback actions, discussed in more detail below, may typically include at least one type of action and, optionally, one or more variables that characterize the performance of the action. Thus, for example, a change in vaporization temperature would be a type of action, and an increase or decrease or an amount of increase or decrease would represent a variable that characterizes the performance of the action. Similarly, a modification of the active ingredient concentration in the aerosol would be a type of action, and an increase or decrease or an amount of increase or decrease in concentration would represent a variable that characterizes the performance of the action.

[0236] Thus, as a non-limiting example in the context of a machine learning system, different output nodes may represent different types of behavior, and the values ​​of these nodes may represent flags indicating the selection of that feedback behavior or values ​​for variables of that feedback behavior depending on how the system is trained. It should also be appreciated that in a machine learning system, multiple output nodes may be associated with one or more types of behavior depending on the training format.

[0237] Of course, potentially multiple feedback actions may be indicated in response to the estimated user state. In such a situation, the feedback processor may then decide whether to select only one feedback action or to execute multiple feedback actions in parallel or sequentially, for example based on the degree of change caused by the action implied by the associated variable or variables, optionally ordered according to a predefined order, also in response to the strength of activation of the flag output node for each feedback action and / or the degree of change implied by the associated variable or variables for each action.

[0238] It will also be appreciated that for training such machine learning systems, measured and / or reported user states can be provided as inputs and respective suggested feedback actions provided as targets, with actions and values ​​being selected according to their reported effects in user trials for users with the corresponding user states, again with the effect or efficacy typically relating to an improvement in a user-perceived state and / or a change in neurological and / or physiological state towards a predefined baseline or preferred state.

[0239] Optionally, the use of simulated conditions and corresponding feedback actions as a first training phase can provide initial training (e.g., based on survey results as described above), after which the model can be refined using a relatively small cohort of real-world training data.

[0240] Optionally, the user's own feedback regarding the effectiveness and / or suitability, desirability, usefulness, etc., of any feedback actions may be further used to refine the model and effectively customize it for the user. This feedback may likewise be reported by the user based on measurements of the user interface of devices in the delivery ecosystem, such as the delivery device or phone, and / or neurological and / or physiological responses. If multiple feedback actions are performed or specified, the user may optionally rank each in order of preference.

[0241] In summary, a two-stage process that includes explicit estimation of user state as a first or interim stage, whether performed by rule-based techniques or machine learning, may be used when these stages are a better fit to the available underlying empirical data sets used to model the correspondence / correlation.

[0242] Objectively, in this mode, the estimation processor operates as described above to take inputs from the acquisition processor, typically in the form of different individual, subset, and / or class-level user factors, and output one or more suggested feedback actions, including simply identifying the actions in a manner similar to a flag, identifying the relevance of the actions to the estimated state based on the activation levels and outputs corresponding to the suggested feedback actions, and / or specifying a change or amount of change in one or more variables that at least partially characterize the suggested feedback action.

[0243] Thus, the explicit estimation of the user state is usually an internal interim step. However, it can of course also be relayed as information to the user, who can optionally modify the estimate, especially if the estimate or a component of the estimate in the multivariate description relates to a subjective measure or a surrogate of a subjective measure such as the user's sense of stress. Thus, the estimate can be displayed, for example, in a user interface of the user's mobile phone, and the user can use this information to perform a self-assessment and change the estimate as a result. The revised estimate of the user state can then be used in a second step in addition to or instead of the initially generated estimate to generate a proposed feedback action that is considered more accurate than a proposal based on the original estimate of the user state.

[0244] Furthermore, any changes made to the user's state estimate can be used to update and refine the first stage model; indeed, for certain machine learning techniques, the absence of user corrections can likewise be seen as a positive reinforcement of the estimate for training purposes.

[0245] As discussed earlier in this specification, if further training is not desired, the relationships between input and output values ​​resulting from the machine learning process may optionally be captured in one or more lookup tables, which may be computationally easier to use (although potentially occupy more memory).

[0246] Implicit State Estimation In one embodiment herein, rather than using the two-stage process described above, the estimation processor performs a one-stage process that implicitly estimates the user's state as part of the relationship between the individual, subset, and / or class-level user factors provided as input by the acquisition processor and the suggested feedback actions generated as output, typically expected to change the user's state.

[0247] Thus, an estimation processor (1020) configured to calculate an estimate of a user state based on one or more of the obtained user factors may be equivalent to an estimation processor (1020) configured to generate a suggested feedback action state based on one or more of the obtained user factors, where the user state is implicit in the relationship between the user factors and the suggested feedback action expected to change the user's implicit estimated state.

[0248] Similar to the two-stage process described herein above, the estimation processor may convert the input data from the acquisition processor into an estimated state through the use of predefined rules, algorithms, and / or heuristics, which may, for example, combine the two separate stage processes of the explicit state estimation embodiment, and / or refine some or all of the rules, diagrams, and / or heuristics in response to the one-stage nature of the implicit state estimation approach, or may be derived from scratch in the case of a one-stage process.

[0249] Thus, in one embodiment of the present specification, a decision tree (e.g., the decision tree mentioned in connection with the first stage of the inference processor) may be used in a similar manner as above. In this case, each path of the decision tree similarly represents a (usually unique or uniquely ordered) met criterion, and thus a threshold number of met criteria representing the possibility of stress or high arousal (or potentially other states depending on the selection criteria or sequence of criteria). The number of criteria may be different as required for different paths. As a result, the terminal nodes of each path may correspond to a particular feedback action (or a selection of suitable feedback actions to be selected separately) appropriate for the set / sequence of met criteria. In this form, the decision tree may embody a one-stage approach to capture met criteria and provide feedback, with the user's state being implicit in the structure of the decision tree and the feedback action selected at each terminal node.

[0250] Again, as with the two-stage process, the estimation processor may alternatively or additionally use look-up tables to convert the input data into suggested feedback actions, which may similarly be a concatenation of look-up tables from the two-stage approach and / or further processed to provide a one-stage look-up table, or derived from scratch in the case of the one-stage process.

[0251] Again, as with the two-stage process, the estimation processor may alternatively or additionally use machine learning, for example by using inputs used in a first stage of explicit state estimation and goals used in a second stage of generating suggested feedback actions from the estimation stage to train a machine learning system that identifies a measurable correspondence between the two.

[0252] Of course, to present corresponding inputs and goals for training, the training set must capture this correspondence. As described above in this specification, there may be data sets for inputs and user states, as well as user states and effective feedback actions. As a result, inputs and feedback actions can be combined for training purposes, if necessary based on common user state values, classes, or multivariate descriptors. Obviously, if the training data set is collected by users who have measured and / or self-reported user factors, measured and / or self-reported user states, and measured and / or self-reported the effectiveness, suitability, desirability, usefulness, etc. of subsequent feedback actions, then a self-consistent set of input user factors and goal feedback actions (as provided by the acquisition processor) can be used for training.

[0253] Alternatively or in addition to the above, a two-stage system of explicit state estimation trained on separate data sets, a two-stage system of explicit state estimation using each rule, algorithm, and / or heuristic from the two stages, and / or a two-stage system of explicit state estimation using lookup tables from the two stages can be used as data sources.

[0254] For example, a lookup table for two-stage estimation or creation of a one-stage lookup table by running rules, algorithms, and / or heuristics and / or machine learning systems through the first and second stages may provide a lookup link between inputs as provided by the acquisition processor and the suggested feedback actions generated by running the two-stage process using those inputs.

[0255] Alternatively or in addition to the above, lookup tables or rules, algorithms, and / or heuristics for two-stage estimation and / or training of a one-stage machine learning system by running the machine learning system through the first and second stages may provide inputs such as those provided by the acquisition processor and provide goals for training the suggested feedback actions generated by running the two-stage process using those inputs.

[0256] Optionally, a single-stage machine learning system so trained may then refine its training using additional data, such as a composite training set as described above, and / or data received from one or more users during use of a user feedback system, in a manner similar to that described herein above for the staged approach.

[0257] It should also be appreciated that, for example, a training set may be based directly on the incorporation of desired input and target values ​​rather than using an amalgamation of data sets or processes.

[0258] Of course, for either the two-stage or one-stage approach, a training set may be constructed, for example relating user factors to user states, by collecting training data using one or more devices in the delivery ecosystem. Such a training set may be generated by a version of the user feedback system that does not generate suggested feedback actions, but simply collects user factor and user state information. Similarly, a training set relating user states to suggested feedback actions may initially be based on questioning users whose respective states are known (e.g., measured / reported) to evaluate suggested feedback actions via a phone user interface, for example as part of a user testing regime. Thus, in this case, the feedback system may suggest feedback actions and select one or more of the suggested actions, but in a different version or mode, may present the selected suggested feedback action(s) to the user (e.g., via a user interface) for evaluation, for example during a training data collection phase or calibration phase (e.g., characterizing users in subgroups that can better tailor responses, as disclosed elsewhere herein), and / or modify or prompt the user to modify one or more operations of at least a first device in the delivery ecosystem by executing the selected suggested feedback action(s) in response to an estimation of the user state (whether explicitly or implicitly modeled) in a manner that is expected to change the estimated state of the user. Training data relating user factors to suggested feedback actions may be obtained as well.

[0259] Such data sets may therefore be obtained using a version or mode of a user feedback system that does not actually result in modification of one or more operations of devices in the ecosystem (optionally other than to elicit a response from the user, e.g., for training data purposes), as described above.

[0260] Thus, such a pre-generation of a user feedback system or training / refinement mode of the user feedback system may include an acquisition processor (1010) operable to acquire one or more user factors indicative of a user state, and to acquire user state data (e.g., based on measurements similar to the user factors and / or self-report by the user) and / or feedback action preference / effect data. The estimation processor would then include a training or development phase, e.g., once a sufficient corpus of data has been accumulated, in which correspondences / relationships / correlations between inputs based on user factors as described above and goals based on user state (in a two-stage approach) or suggested feedback actions (in a one-stage approach) are modeled as described above.

[0261] Alternatively or in addition to the above, in a pre-generation and / or training mode of such a feedback system, the delivery device and / or other participating devices in the delivery ecosystem may result in only uploading data to the acquisition processor and not downloading feedback operations (or optionally any other data) from the feedback system.

[0262] Similarly, pre-generation and / or training modes of such feedback systems and / or providing refinement or supplemental input to the feedback system during normal use may involve direct input of the user's state as reported by the user to user factors such as from neurological / physiological data (e.g., via biosensing), movement and / or location user factors (e.g., via touch, accelerometer, or GPS sensors), contextual user factors, and / or any of the other user factors disclosed herein, as described elsewhere herein, which may be used to generate a training set as described above, but alternatively or additionally, the user's reported state may be processed as a user factor directly by the acquisition or estimation processor.

[0263] The user's reported state may be obtained based on the results of a questionnaire presented to the user via a user interface on a device in the delivery ecosystem, such as the user's mobile phone or delivery device. Such a questionnaire may simply ask the user to select, for example from a menu or drop-down box, a state that most closely matches their subjective understanding of the respective state, and / or ask questions whose answer selection by the user may indicate the user's state. Such a questionnaire may be provided periodically or in response to some event, such as a detected user factor or a value associated with the user factor reaching a certain threshold. Thus, for example, a questionnaire may be provided in response to a certain calendar event, the user visiting a certain location, or the user's heart rate exceeding a predefined threshold. Optionally, the frequency with which the questionnaire is provided to the user may be limited so as not to burden the user with the questionnaire process. Similarly, the extent of the questionnaire may be optionally modified to be more of a direct self-assessment than an assessment based on responses to questions such as those described above (if provided).

[0264] More generally, as with other questionnaires described herein, the user may be optionally presented with, offered to, or required to complete the questionnaire depending on many factors. For example, a questionnaire regarding the user's cultural background may be conducted once, e.g., as part of an initial setup process, and may optionally require confirmation, e.g., annually. Other questionnaires, on the other hand, may be periodic, e.g., monthly or weekly, and may passively wait for updates by the user, such as the user's current weight. Other questionnaires may be event-based, e.g., in response to detection of deviations in puff frequency or behavior, or feedback intervention. In these cases, there is optionally a timeout that does not conduct the same or other questionnaires within a predefined period (e.g., 1-N hours or 1 day) to avoid burdening the user. Optionally, if the user agrees to more intensive questioning (e.g., as part of a system training program), the questionnaires may be conducted more regularly (e.g., every few hours) and when the user is awake.

[0265] In principle, the user's reported state may optionally be used as a substitute for an explicit state estimate by the estimation processor, but in at least some cases it may be an approximation compared to what can be derived or estimated from some measurements (if available) and the user is not informed of all the facts that may be available to the feedback system. Furthermore, some users may normalize and preconceive their state, especially in the case of pathological conditions such as depression, and self-report with a preconceived idea. For this reason, the user's direct input of the state may optionally be used in the first stage (or only in the first stage) as input to the estimation processor, together with one or more other user factors from the acquisition processor as described above. Optionally, alternatively or additionally, in the case of using a two-stage technique, the user's direct input of the state may be used as input to the second stage of the estimation processor, together with the estimate of the respective state.

[0266] Other variables in the training and inputs are also contemplated. For example, it will be appreciated that, as discussed herein above, different user factors will operate or vary over different time frames. As a result, in the case of a two-stage or one-stage approach to the estimation processor as described herein, user factors that are not expected to change within the interval between successive operations of the estimation processor may be stored (e.g., in storage 1012) and reused, rather than being reacquired.

[0267] Furthermore, some of the estimation models for these long-term factors may not need to be re-run if the results of these factors are not expected to change. This may be easy for rules, algorithms, and / or heuristics and / or look-up tables, but may require architectural changes for machine learning systems. For example, a two-stage ML or multi-layer system may be trained on all inputs, but then run with the inputs or outputs for the long-term user factors clamped, and the computed intermediate results of that part of the ML system may be fed to other parts of the ML system along with newly generated intermediate results from the user factors in a shorter time frame.

[0268] It is also understood that, as described herein above, different users may have different combinations of devices in their delivery ecosystems and / or different combinations of these devices may be active at any one time. Similarly, different users may have a larger or smaller social media presence, or may use their digital calendars to a larger or smaller extent. As a result, the user factors available to the acquisition processor and thus the inputs available to the estimation processor may vary from user to user and / or from time to time. The estimation processor may therefore suggest feedback actions by using different models (explicit or implicit, as described above) depending on the available inputs. Alternatively or additionally, if an input to a model is missing, a neutral input value may be provided to reduce or eliminate the impact of the missing input on the suggested feedback action. The number of different models provided by / to the estimation processor may therefore depend on the number of data sources assumed by the model (potentially more or more diverse data sources would make the model more vulnerable) and the robustness of the model to the replacement of an input with a placebo / neutral value when the input is currently unavailable. In the latter case, it is understood that some inputs may be deemed more important than others and therefore at least some individual inputs may be required to run the model. Thus, depending on the complexity and robustness of the model, only one model may be required or a number of models may be required for different scenarios. Optionally, a subset of all available models may be selected for a user according to the devices known to be in the delivery ecosystem. However, new models may be added when new devices join the delivery ecosystem, whether permanently, e.g., when the user purchases a new dock 200, or temporarily, e.g., when the user interacts with a vending machine or point-of-sale device.

[0269] Estimated Processor Output Whether a one-stage or two-stage process is used, and whether the estimation of any stage is based on rules, algorithms, and / or heuristics, look-up tables, and / or machine learning, the output of the estimation processor is a suggested feedback action.

[0270] The possible feedback actions differ qualitatively and / or quantitatively.

[0271] Thus, for example, feedback actions may vary qualitatively based on modifying aerosol generation for the user (whether in response to or in anticipation of the current situation), modifying the user's interaction with the delivery device or system during or between inhalations, modifying the user interface of the delivery device or system, reminding the user to use or modify the use of the delivery device or system, recommending the operation or selection of a delivery device or delivery device consumables, and / or recommending / activating / modifying an operation of a device that is not directly related to the delivery of the active ingredient, but which can directly change the user's condition (e.g., through biofeedback) or indirectly change it (e.g., by activating noise cancellation on the user's headphones).

[0272] Thus, more generally, feedback actions may be categorized into behavioral categories that focus on altering a user's behavior and / or habits to alter their state, pharmaceutical categories that focus on the manner in which one or more active ingredients delivered to the user alter their respective state, and non-consumable intervention categories that focus on alternative first or third party options for altering the user's state (i.e., associated with the delivery device, other devices in the delivery ecosystem, or elsewhere).

[0273] On the other hand, suggested feedback actions may vary qualitatively depending on the degree to which it is desired that the effect of the feedback action will bring about a positive change in the user's condition. Thus, for example, in a delivery device, changes in heater temperature, payload aerosolization, payload composition, etc. may include a quantitative value indicating the degree of change or class of change, as appropriate. Similarly, modification of the user interface in the delivery device or another device in the delivery ecosystem may include incremental steps regarding the number of user interactions required or prompted with the delivery system as well as the nature of these user interactions. For example, it may be performed through five categories, the first category having no notifications to minimize user interruption, the second category having only important notifications such as low battery or low payload, the third category corresponding to a default where important and non-important notifications are provided, the fourth category further including recommendations and / or prompts to have the user have other features of the user interface, and the fifth category additionally including an audible tone. These five categories may be selected on a scale of stress (e.g., minimal notifications when stressed) and / or boredom (e.g., loud notifications when bored) depending on the user's state.

[0274] As discussed above, the type of feedback action and / or the amount or class of change may be identified according to rules, algorithms, and / or heuristics, look-up tables, and / or machine learning, as appropriate.

[0275] Similarly, as mentioned above, if multiple types of feedback actions and / or multiple change amounts or change classes are calculated / estimated as appropriate responses to a user factor / user state, then optionally multiple feedback actions may be suggested accordingly, or the top N feedback actions may be selected, for example based on activation intensity, where N may be 1 or more.

[0276] Feedback Processor The feedback processor 1030 is operable to modify one or more operations of devices in the delivery ecosystem in response to the estimation of the user state through execution of one or more suggested feedback actions.

[0277] The feedback processor may therefore act to appropriately cause one or more feedback actions suggested by the estimation processor to occur within the delivery ecosystem.

[0278] The feedback action(s) are typically performed in a manner that is expected to change the estimated state of a user, who is considered to be a typical, average, expected user. Of course, the model(s) underlying the generation of the suggested feedback action(s) are typically developed or trained using data from a corpus of users, and therefore relate to changes in the state of a typical, average, or expected user.

[0279] However, most users are likely to react similarly to these changes, so typically the conditions for a particular user of each delivery device will change in a similar manner.

[0280] However, as described elsewhere herein, if the feedback system can receive separate feedback from individual users (e.g., by measurement or self-report) regarding the effectiveness of the proposed feedback action, then the feedback action can optionally be performed in response to the estimation of the user state in a manner that is expected to alter the estimated state of the particular user, e.g., by supplemental training and / or parameter refinement, allowing further adjustments for the particular user. Similarly, separate rules, algorithms, and / or heuristics, lookup tables, or machine learning systems may be generated for different groups of users, e.g., based on attributes and / or patterns of responses to the feedback action, such that the proposed feedback action is further tailored to a particular user in one of these groups, even if no measured or reported evaluation of the effectiveness of the feedback is available from a particular user, even if the machine learning system is too small to effectively refine its training or change parameters of the algorithm, etc. to customize the response.

[0281] Like the acquisition and estimation processors, the feedback processor 1030 may comprise one or more physical and / or virtual processors and may be located in a remote server 1000 and / or functionality may be distributed or otherwise distributed across multiple devices in the delivery ecosystem, including but not limited to the user's mobile phone 100, the docking unit 200, the vending machine 300, and the delivery device 10 itself. The feedback processor may comprise one or more communication inputs, for example for receiving data from the estimation processor 1010, and one or more communication outputs, for example for communicating with the delivery device 10 and / or another device in the delivery ecosystem 1 as listed above, or any other device that may participate in the feedback operation.

[0282] In particular, the feedback processor may optionally be located on a server and / or a device in the delivery ecosystem having suitable computing capabilities, such as a vending machine or a mobile phone, and may comprise a selection and notification sub-processor (not shown) that is optionally capable of selecting one or more feedback actions and selecting one or more respective devices in the ecosystem to execute the one or more feedback actions, and optionally an action execution sub-processor (not shown) that manages the execution of the feedback actions on one or more respective devices in the ecosystem. Optionally, the action execution sub-processor may be considered a separate processor from the feedback processor.

[0283] References herein to the selection and notification sub-processor and the feedback processor, or to the action performing sub-processor and the feedback processor, respectively, are to be considered as equivalent. It will be appreciated that these sub-processors may be complementary hardware to the feedback processor and / or may effectively share the role of the feedback processor, while also being equivalent in functionality to the feedback processor operating under suitable software instructions. However, as noted above, at least the action performing sub-processor may optionally be a processor separate from the feedback processor, communicating with the feedback processor, for example via the internet.

[0284] Choices and Notices Optionally, the selection and notification sub-processor may select one or more of the feedback actions generated by the inference processor as described herein above if the inference processor indicates that more than one feedback action may be appropriate (e.g. when a set of satisfied criteria is associated with a selection of feedback actions, when a decision tree node is associated with a selection of feedback actions, when a machine learning system indicates multiple feedback actions of differing activation strengths, or when a rule or heuristic allows multiple feedback actions as options, etc.). Obviously, if only one feedback action is proposed, this will be selected as the default.

[0285] For the selected feedback action, the selection and notification sub-processor may then select one or more devices in the delivery ecosystem on which the feedback action should be performed, and formulate a command / notification / instruction for the or each device characterizing the type and / or amount of the feedback action. Of course, if only one feedback action is possible for a device, the type may be implicit in the notification act, and similarly, if only one amount of feedback action is possible for a device, the amount may be implicit in the notification act. Any device in the delivery ecosystem may potentially comprise a feedback means. Thus, of course, the device or devices providing user factor data to the acquisition processor in the delivery ecosystem may potentially be different from the device or devices performing the or each feedback action.

[0286] Optionally, the selection and notification sub-processor may poll devices in the delivery ecosystem to determine their availability for purposes of providing feedback actions, e.g., for devices accessible by the processor via the internet, devices registered in association with the user or the user's delivery device (e.g., delivery device 10, mobile phone 100, wearable device 400, docking device 200) may be polled directly.

[0287] For devices that are only accessible via an intermediate device (e.g., a Bluetooth connection to the accessible device), the accessible device may be required to poll such indirect devices. Thus, for example, the selection and notification sub-processor may cause the user's mobile phone 100 to poll / request the delivery device 10, the wearable device 400, or the docking device 200 (if only accessible via a local wired or wireless connection).

[0288] For devices that are not formally associated with a user, such as a vending machine 300 or other point of sale system, or that are only intermittently associated with a user, the selection and notification sub-processor may receive location data from devices in the delivery ecosystem associated with the user, such as a mobile phone 100 or delivery device 10, and compare it to the registered or reported location of the vending machine 300. If the locations are within a threshold distance of each other, the vending machine is considered part of the delivery ecosystem while that condition applies. Alternatively or additionally, the selection and notification sub-processor may instruct the accessible device to poll any compatible vending machines or to broadcast a Bluetooth beacon that identifies the accessible device, for example by using a disposable ID to allow the accessible device to be identified without revealing details of the user or its associated device. Such an ID may include a component that identifies the purpose of the ID to allow detection by the vending machine, followed by a disposable component that is unique to the user or its associated device. A compatible vending machine according to embodiments of the present invention may then optionally recognize and relay the single-use ID to a selection and notification sub-processor, thereby notifying the user that they have an accessible device within local wireless range of the vending machine. Of course, although a vending machine has been mentioned above, this is only one illustrative example and these techniques may also apply to any device that is not formally associated with a user or that is only intermittently associated with a user, such as a car or train, a Wi-Fi or Bluetooth hotspot in a store, a smart TV, etc.

[0289] Optionally, devices outside the user's own delivery ecosystem may be selected. For example, friends or family may be allowed to intervene by notifying them of the user's status through the use of delivery devices and / or devices such as phones of friends or family associated with the user (e.g., following registration of these people by the user). Optionally, the user may set the conditions under which this occurs and / or under which friends or family are notified. Similarly, devices within a predefined proximity area of ​​the user may be selected. For example, if the user is feeling good, all compatible devices within a predefined radius of the user may synchronize characteristics such as light color to inform the user that a fun social encounter is occurring.

[0290] By using one or more of these techniques, the selection and notification sub-processor may determine the devices currently available for delivery of feedback actions.

[0291] Typically, feedback actions are specific to a particular device in the delivery ecosystem or to a pair of devices that cooperate to perform a function, so that for a proposed or selected feedback action, the selection and notification sub-processor may optionally only poll the device or devices in the delivery ecosystem that are relevant to that feedback action.

[0292] Similarly, it will be appreciated that certain devices in the delivery ecosystem may provide input data to the feedback system that is used to generate the suggested feedback actions. As a result, input actions from such devices may be recorded as indicative of their respective accessibility and / or the suggested feedback actions may imply that a particular device is currently accessible to the feedback system. In either case, polling of the devices may not be necessary, or if a polling scheme is present, receipt of the input data may be treated as an effective polling result.

[0293] If one or more devices associated with a feedback action are unavailable (e.g. do not respond to polling), the feedback processor / selection and notification sub-processor may optionally select the next suggested feedback action among the top N feedback actions if multiple feedback actions were suggested by the inference processor. If no associated devices are available for a feedback action, the feedback processor may not perform any feedback action and / or send a notification to the user, for example via a user interface on the user's phone, or, if the user's phone is not available as an accessible device for linking to other devices in the ecosystem, notify the user via a text or other similar mechanism to be delivered once the user is contactable again.

[0294] If one or more devices associated with the feedback operation are available (i.e., if they respond to polling, if they respond to polling within a predetermined preceding period during which the device may still be assumed to be accessible, or if they provide input data within a predetermined preceding period), the feedback processor will send one or more commands to the one or more devices to perform the feedback operation as suggested by the estimation processor.

[0295] As mentioned above, the nature of the command may depend on the proposed action as well as the target device or devices. In some cases, the mere presence of a command will be sufficient to specify the proposed action (e.g., turning on a device that is off). In other cases, the command will need to specify the type of feedback action, e.g., with respect to changing heater function in the delivery system, payload type, user interface behavior, etc. In any of these cases, the command may need to specify the amount of feedback action, e.g., changing temperature, concentration of active ingredients or fragrances in the payload, or selected parameters for the user interface.

[0296] As mentioned above, commands may be passed directly to the accessible device, or the accessible device may request that the accessible device relay the command to another device in the ecosystem or itself issue a command to such a device. For example, the feedback processor may instruct the user's mobile phone 100 to issue a command to the delivery device 10. Other degrees of indirection may be envisaged, such as the user's mobile phone issuing a command to the dock 200, which may modify settings of the delivery device when the dock 200 is docked (e.g., for charging or to charge a payload). Similarly, it will be appreciated that the feedback processor may issue different types of commands to different devices. Thus, for example, a command to change an aspect of the user interface may be issued directly to the mobile phone, which in turn may be issued to the dock 200 (directly, if possible, or via a phone call) to cause the dock 200 to change the composition of the payload provided to the delivery device, as well as modify one or more settings of the delivery device when docked. It will be appreciated that other permutations of such commands (whether direct, indirect, or a mixture of the two) are envisaged in the delivery ecosystem.

[0297] As described elsewhere herein, it will be appreciated that various feedback actions may relate to the behavior of the delivery ecosystem, the drug, and / or non-consumable aspects.

[0298] Behavioral feedback actions typically focus on modifying user behavior and habits regarding operation and / or interaction with devices in the delivery ecosystem other than actions related to the amount or nature of the active ingredient delivered by the delivery device itself, but which may occur in parallel. Examples may include modifying the flavor or flavor concentration, modifying inhalation behavior by modifying the delivered vapor volume, modifying scheduling schemes or reminders associated with or correlated to delivery device usage, providing information, modifying the user interface in terms of feedback modes (e.g., haptics and / or visuals such as colored lights, graphic themes, and / or messages) (whether on the delivery device or another device in the delivery ecosystem) (e.g., providing a traffic light UI display on the delivery device, such as LEDs, to alert the user to the mode of device usage), etc.

[0299] Pharmaceutical feedback actions focus on pharmaceutical interventions that change the user's state, typically based on active ingredient such as amount or type, timing of changing these (e.g., as a response or preemption based on correlations between current user factors and future user state or feedback actions), etc. Such actions may also relate to the selection of alternative consumption modes (e.g., switching from vaping to snus or vice versa).

[0300] Non-consumptive feedback actions typically relate to activating / controlling or simply recommending the use of devices not specifically related to the consumption of an active ingredient, such as aromatherapy systems / steamers, biofeedback devices, headphones (e.g. activating noise cancelling or modifying the volume or music selection), vehicle use (e.g. stress warnings, or selecting / reselecting a longer but less congested or slower route), etc.

[0301] The Selection-Notification Sub-Processor may consist of one or more real or virtual processors, and its functionality may be located or distributed within one or more devices in the server and / or delivery ecosystem as appropriate.

[0302] Executing an action The action execution sub-processor may be optional - for example, some devices are capable of accepting commands directly without further interpretation or processing, in which case the action execution sub-processor may not be required and its role may be fulfilled by the feedback processor / selection and notification sub-processor.

[0303] On the other hand, in some cases the role of the action execution sub-processor may actually reside within the device, e.g. capable of interpreting user interface commands and modifying the operation of the device, in which case commands from the feedback processor may optionally simply replicate such user interface commands.

[0304] In other cases, the operation execution sub-processor may be provided separately, for example by employing a conventional processor responsive to suitable software instructions. Such an example may be an app on a mobile phone operable to receive commands and modify one or more aspects of the user's mobile phone and / or apps on the mobile phone, the delivery device, and / or one or more other devices in the delivery ecosystem. Similarly, the delivery device dock 200 may be equipped with such an operation execution sub-processor, as may multiple types of delivery devices.

[0305] The action execution sub-processor is operative to execute a feedback action on the or each associated device. Thus, for example, if a command for a feedback action describes changing a heater temperature of a delivery device, the action execution sub-processor may effectuate the specified change by changing a power source for the heater and / or a duty cycle for the heater.

[0306] Similarly, for example, if a command relating to a feedback action describes reducing environmental noise levels for a user, the action execution sub-processor may cause a pair of noise cancelling headphones to activate a noise cancelling function, and the action execution sub-processor may cause the user's mobile phone to lower the volume of music playing in the headphones and display a message to the user encouraging the user to avoid noise sources in the environment.

[0307] The particular operations that each sub-processor performs may thus depend on the nature of the suggested feedback operation and the nature of the devices within the delivery ecosystem, but will typically represent a direct translation of the suggested feedback operation into mechanisms that can be implemented within the device.

[0308] As described herein above, the feedback action may be accompanied by or followed by a request or opportunity for the user to report on its effectiveness. Alternatively or additionally, the feedback action may be accompanied by or followed by a positive reinforcement of the expected state change, for example through a message on the UI, a color change in the interface, a haptic response, etc. Alternatively, the feedback action may be accompanied by or followed by a positive goal to be achieved in an app for the wearable. The reinforcement may be a simple message indicating that feedback has occurred, or may be based on a measurement to report, for example, that the user's heart rate has slowed, or may be to confirm (usually as evidenced by a change in one or more user factors or as a self-report by the user) that an action (by changing the user's state) has worked well. The recognition and / or expectation of a state change brought about by such positive reinforcement may increase the effectiveness of at least some of the feedback actions.

[0309] The operation performing sub-processors may consist of one or more real or virtual processors, the functionality of which may be located or distributed within a server and / or one or more devices within the delivery ecosystem as appropriate.

[0310] Processor As mentioned above, the acquisition processor, the estimation processor, and the feedback processor (and any sub-processors) may comprise one or more real or virtual processors located in one or more servers and / or delivery ecosystems. Moreover, it should be understood that the division of roles described herein is not fixed. For example, the acquisition processor may receive information directly indicative of the user's state (e.g., by the user's self-report), so that the first stage of the estimation processor's two-stage process can be bypassed or supplemented by the acquisition processor. Similarly, in this case, the feedback processor may, for example, search for a corresponding suggested feedback action. Thus, in this example, the role of the estimation processor is performed by the acquisition processor and the feedback processor. Thus, more generally, these processors are representative of tasks that may be implemented by any processor under suitable software instructions, and are considered to include, equivalently, any of the following tasks: data collection tasks, feedback suggestion tasks (whether or not based on explicit estimation of the user's state), and feedback learning or feedback providing tasks.

[0311] In one embodiment herein, data may be collected from sensors on the delivery device and transmitted to a companion device in the delivery ecosystem, such as the user's phone, as described elsewhere herein. The data may optionally undergo at least some pre-processing by the delivery device prior to transmission. As described elsewhere herein, such sensor data from the delivery device may then be associated with data from other sources (e.g., information related to the user's state before, during, and / or after recording of the sensor data, such as the aforementioned environmental data, deterministic data, contextual data, neurological data, physiological data, indirect data, and historical data) to generate data related to multiple data sources. This data may be sent to a back-end server (e.g., at least a portion of the acquisition server functionality occurs at such a server). Alternatively, the data may be processed by a smartphone acting as an acquisition processor, and the estimation processor functionality may be performed on one of the phone, the back-end server, or a mixture of the two. In this case, the feedback server functionality may similarly be performed on the phone, on the back-end server, or on one of the two hybrids (e.g., depending on the nature of the feedback or follow-on action, actions such as sending notifications or product requests may be performed on the back-end server in one embodiment and on the phone in another). The combination of all three processors implemented on the smartphone has the advantage of user privacy, since data is not sent to the back-end server. Alternatively, the inference processor functionality may be performed by the server, for example, since it may be computationally complex. On the other hand, the input to the inference processor may only be machine readable and may optionally be associated with an anonymized request (e.g., a single user request number), allowing processing of the data without the back-end server user being able to determine the specific user's input or inferred state.

[0312] Overview of embodiment In one general embodiment herein, a user feedback system for a user of a delivery device (10) in a delivery ecosystem (1) comprises: first, an acquisition processor (1010) configured to acquire one or more user factors indicative of a state of the user, as described elsewhere herein; and second, an inference processor (1020) configured to identify a one- or two-step correspondence or correlation between the one or more user factors indicative of the acquired user state and at least a first feedback action, as described elsewhere herein, the feedback action being expected to change the state of the user as at least partially indicated by the one or more user factors, as described elsewhere herein (e.g., where the user factors at least implicitly indicate a state, such that a correlation between the two can be derived).

[0313] In one example of this summary embodiment, the two-stage correlation includes a first correspondence or correlation between one or more user factors indicative of an acquired user state and at least a first state of the user, as described elsewhere herein, and a second correspondence or correlation between the at least first state of the user and at least a first feedback action, also as described elsewhere herein.

[0314] In this example, the estimation processor is optionally operable to calculate an estimate of at least a first state of the user based on the obtained one or more user factors by using a model including correspondence or correlation data between the one or more user factors and one or more user states, as described elsewhere herein.

[0315] In this case, the model optionally embodies the corresponding data in one or more selected from the list consisting of one or more lookup tables that associate one or more user factors with one or more user states, as described elsewhere herein, and one or more heuristics (e.g., algorithms) that use one or more user factors as inputs to generate one or more user states as outputs, as described elsewhere herein.

[0316] Similarly, in this case, the model optionally embodies the correlation data in one or more machine learning models that use one or more user factors as inputs to generate one or more user states as outputs, as described elsewhere herein.

[0317] Here, the machine learning model is optionally trained based on inputs including one or more selected from the list consisting of one or more individual values ​​each based on one or more user factors, one or more combined values ​​each based on two or more user factors, and one or more values ​​each based on user factors from a single data class, as described elsewhere herein.

[0318] Also herein, the machine learning model is optionally trained based on inputs including self-reported estimates of the user's state obtained from the user, as described elsewhere herein.

[0319] Also herein, the machine learning model is optionally trained based on a target output user state including one or more selected from the list consisting of a self-reported estimate of the user state obtained from the user, an estimate of the user state obtained by separate analysis of user factors, and an estimate of the user state derived from the self-reported estimate obtained from the user with respect to the effect of the identified at least first feedback action, as described elsewhere in this specification.

[0320] Also herein, the machine learning model is optionally trained based on target output user states formatted as one or more selected from the list consisting of single representative values, representative categories, and multivariate representations, as described elsewhere in this specification.

[0321] However, in this example, the estimation processor is optionally configured to calculate an estimate of the user's first state based on whether a threshold number of user factors indicative of at least the first state of the user satisfy respective predetermined criteria, as described elsewhere herein.

[0322] In this case, the number of user factors indicating the first state of the user that meet each predetermined criterion optionally indicates the likelihood that the user is in the first user state, as described elsewhere in this specification.

[0323] Similarly, optionally in this case a subset of user factors indicative of the first state of the user is selected, as described elsewhere herein, and the threshold number is part of that subset.

[0324] Here, the subset of user factors is optionally selected based at least in part on their current availability, as described elsewhere herein.

[0325] However, in this example, the estimation processor is optionally configured to calculate an estimate of the user's first state by using one or more decision trees including branches corresponding to multiple sets or multiple subsets of user factors indicative of at least a first state of the user satisfying respective predetermined criteria, as described elsewhere herein, and leaves of the decision trees corresponding to an indication of whether the user is at least in the first state.

[0326] In this case, optionally, as described elsewhere in this specification, for leaves that indicate the user is in the first state, the length of each branch of the decision tree indicates the likelihood that the user is in the first state.

[0327] However, in this example, the estimation processor is optionally operable to identify at least a first feedback action (e.g., calculate / generate an output corresponding to the feedback action) based on the calculated estimate of the user state by using a model including correspondence or correlation data between one or more user states and one or more feedback actions, as described elsewhere herein.

[0328] In this case, the model optionally embodies the correspondence or correlation data in one or more selected from the list consisting of one or more lookup tables that associate one or more user states with one or more feedback actions, and one or more heuristics that use one or more user states as inputs to identify one or more feedback actions as outputs, as described elsewhere herein.

[0329] Similarly, in this case, the estimation processor is optionally configured to identify one or more feedback actions associated with a set including a threshold number of user factors indicative of a first state of the user that satisfies respective predetermined criteria, as described elsewhere herein.

[0330] Similarly, in this case, the inference processor is optionally configured to identify one or more feedback actions associated with leaf nodes of one or more decision trees including branches corresponding to multiple sets or multiple subsets of user factors indicative of a first state of the user satisfying respective predetermined criteria, as described elsewhere herein.

[0331] Similarly, in this case, the model optionally embodies the correlation data in one or more machine learning models that use one or more user states as inputs to identify one or more feedback actions as outputs, as described elsewhere herein.

[0332] Here, the machine learning model is optionally trained based on input including user states formatted as one or more selected from the list of single representative values, representative categories, and multivariate representations, as described elsewhere herein.

[0333] Also herein, the machine learning model is optionally trained based on inputs including self-reported estimates of the user's state obtained from the user, as described elsewhere herein.

[0334] Also herein, the machine learning model is optionally trained based on inputs that also include at least a subset of user factors, as described elsewhere herein.

[0335] Similarly, in this case, the inference processor is optionally operable to identify one or more suggested feedback actions for one or more selected from the list consisting of behavioral feedback actions affecting at least a first behavior of the user, drug feedback actions affecting consumption of the active ingredient by the user, and non-consummative feedback actions affecting one or more non-consummative actions of the delivery ecosystem, as described elsewhere herein.

[0336] In one example of this summary embodiment, the one-stage correspondence or correlation includes a first correspondence or correlation between one or more user factors indicative of an acquired user state and at least a first feedback action, as described elsewhere herein.

[0337] In this example, the estimation processor is optionally operable to identify at least a first feedback action based on the acquired one or more user factors by using a model including correspondence or correlation data between the one or more feedback actions and the acquired one or more user factors (e.g., by generating / calculating an output corresponding to the feedback action), as described elsewhere herein.

[0338] In this case, the model optionally embodies the correspondence or correlation data in one or more selected from the list of: one or more lookup tables that associate one or more user factors with one or more feedback actions, and one or more heuristics that use one or more user factors as inputs to identify one or more feedback actions as outputs, as described elsewhere herein.

[0339] Similarly, in this case, the model optionally embodies the correspondence or correlation data in one or more machine learning models that use one or more user factors as inputs to identify one or more feedback actions as outputs, as described elsewhere herein.

[0340] Here, the machine learning model is optionally trained based on inputs including one or more selected from the list consisting of one or more individual values ​​each based on one or more user factors, one or more combined values ​​each based on two or more user factors, and one or more values ​​each based on user factors from a single data class, as described elsewhere herein.

[0341] Also herein, the machine learning model is optionally trained based on inputs including self-reported estimates of the user's state obtained from the user, as described elsewhere herein.

[0342] Also herein, the machine learning model is optionally trained based on target output feedback actions including one or more selected from the list consisting of behavioral feedback actions affecting at least a first behavior of the user, drug feedback actions affecting consumption of the active ingredient by the user, and non-consummatory feedback actions affecting one or more non-consummatory actions of the delivery ecosystem, as described elsewhere in this specification.

[0343] However, in this example, the estimation processor is optionally configured to identify one or more feedback actions for the user's first state in response to a threshold number of user factors indicative of the user's first state satisfying respective predetermined criteria, as described elsewhere in this specification, and to identify one or more feedback actions associated with a set including the threshold number of user factors that satisfy respective predetermined criteria.

[0344] Similarly, in this example, the estimation processor is optionally configured to identify one or more feedback actions for the user's first state by using one or more decision trees including branches corresponding to multiple sets or multiple subsets of user factors indicative of the user's first state satisfying respective predetermined criteria, as described elsewhere herein, and each leaf of the decision tree is associated with one or more feedback actions.

[0345] In one example of this general embodiment, the one or more user factors are each based on one selected from the list consisting of at least a first physical characteristic associated with at least a first user inhalation action, at least a first physical characteristic associated with a user behavior other than inhalation, at least a first physical characteristic associated with a user physiology other than inhalation related, and at least a first aspect of the user's situation separate from handling or operation of the delivery device, as described elsewhere herein.

[0346] In one example of this general embodiment, the one or more user factors are each associated with at least one class selected from the list consisting of historical data providing background information about the user, neurological data about the user, physiological data about the user, contextual data about the user, and environmental data about the user, as described elsewhere herein.

[0347] In one example of this general embodiment, the user feedback system comprises a feedback processor (1030) configured to select at least a first feedback action identified by the inference processor for at least a first device in the delivery ecosystem, as described elsewhere herein.

[0348] In this example, the feedback processor is optionally configured to modify one or more operations of at least a first device in the delivery ecosystem according to the selected feedback action or actions, as described elsewhere herein.

[0349] In this case, the device in the delivery ecosystem whose operation or operations are modified is optionally a delivery device (10), as described elsewhere herein.

[0350] In one instance of this summary embodiment, the selected feedback action includes prompting the user to provide feedback regarding the estimated user state to the user feedback system, as described elsewhere herein.

[0351] In one example of this summary embodiment, the selected feedback action also includes prompting the user to provide feedback regarding the selected feedback action to the user feedback system, as described elsewhere herein.

[0352] In one example of this summary embodiment, the feedback processor (1030) is configured to select at least a first identified feedback action in response to a current availability of each device in the delivery ecosystem for execution of the feedback action, as described elsewhere herein.

[0353] In one instance of this general embodiment, the feedback processor is configured to automatically cause at least a first identified feedback action to be performed, as described elsewhere herein.

[0354] In one example of this general embodiment, the feedback processor is configured to prompt the user for consent to have at least some of the identified at least first feedback actions performed, as described elsewhere herein, and to cause only at least some of the identified at least first feedback actions to be performed if consent is determined. Of course, such consent may itself be indicative of the welcomeness or appropriateness of the feedback actions to the user at that time, and thus may be used as input for training a machine learning system.

[0355] In one example of this summary embodiment, the delivery ecosystem comprises one or more selected from the list consisting of one or more delivery devices (10), one or more mobile terminals (100), one or more wearable devices (400), and one or more docking units (200) for the delivery devices, as described elsewhere herein.

[0356] In one example of the present general embodiment, at least a portion of the functionality of one or more of the acquisition processor, estimation processor, and feedback processor, as described elsewhere herein, is provided by a remote server (1000).

[0357] In one example of this general embodiment, at least a portion of the functionality of one or more of the acquisition processor, estimation processor, and feedback processor are provided by one or more processors located within one or more devices (10, 100, 200, 300, 400) of the delivery ecosystem (1), as described elsewhere herein.

[0358] Referring now to FIG. 7, in one general embodiment herein, a method for user feedback to a user of a delivery device (10) in a delivery ecosystem (1) includes the following steps.

[0359] A first obtaining step s710 involves obtaining one or more user factors indicative of a state of the user, as described elsewhere herein.

[0360] The second estimation step s720 includes identifying a one- or two-stage correspondence or correlation between one or more user factors indicative of the obtained user state and at least a first feedback action, as described elsewhere herein, which is expected to change the user's state as at least partially indicated by the one or more user factors, as described elsewhere herein (e.g., where the user factors at least implicitly indicate the state, such that a correlation between the two can be derived).

[0361] Variations of the above methods corresponding to the operation of various embodiments of the methods and / or apparatus as described and claimed herein are considered to be within the scope of the present disclosure and will be apparent to those of skill in the art, including but not limited to the following:

[0362] The two-stage correspondence or correlation includes a first correspondence or correlation between one or more user factors indicative of an acquired user state and at least a first state of the user, as described elsewhere herein, and a second correspondence or correlation between the at least first state of the user and at least a first feedback action, also as described elsewhere herein.

[0363] In this example, the estimation step optionally includes calculating an estimate of at least a first state of the user based on the acquired one or more user factors by using a model including correspondence or correlation data between one or more user factors and one or more user states, as described elsewhere in this specification.

[0364] In this case, the model optionally embodies the correspondence or correlation data in one or more selected from the list consisting of: one or more lookup tables that associate one or more user factors with one or more user states, as described elsewhere herein; and one or more heuristics (e.g., algorithms) that use one or more user factors as inputs to generate one or more user states as outputs, as described elsewhere herein.

[0365] Similarly, in this case, the model optionally embodies the correspondence or correlation data in one or more machine learning models that use one or more user factors as inputs to generate one or more user states as outputs, as described elsewhere herein.

[0366] Here, the machine learning model is optionally trained based on inputs including one or more selected from the list consisting of one or more individual values ​​each based on one or more user factors, one or more combined values ​​each based on two or more user factors, and one or more values ​​each based on user factors from a single data class, as described elsewhere herein.

[0367] Also herein, the machine learning model is optionally trained based on inputs including self-reported estimates of the user's state obtained from the user, as described elsewhere herein.

[0368] Also herein, the machine learning model is optionally trained based on a target output user state including one or more selected from the list consisting of a self-reported estimate of the user state obtained from the user, an estimate of the user state obtained by separate analysis of user factors, and an estimate of the user state derived from the self-reported estimate obtained from the user with respect to the effect of the identified at least first feedback action, as described elsewhere in this specification.

[0369] Also herein, the machine learning model is optionally trained based on target output user states formatted as one or more selected from the list consisting of single representative values, representative categories, and multivariate representations, as described elsewhere in this specification.

[0370] However, in this example, the estimation step optionally includes calculating an estimate of the user's first state based on whether a threshold number of user factors indicative of at least the first state of the user satisfy respective predetermined criteria, as described elsewhere herein.

[0371] Similarly, in this example, the estimation step optionally includes calculating an estimate of the user's first state by using one or more decision trees including branches corresponding to multiple sets or subsets of user factors indicative of at least the first state of the user satisfying respective predetermined criteria, as described elsewhere herein, and leaves of the decision trees corresponding to an indication of whether the user is at least in the first state.

[0372] Similarly, in this example, the estimating step optionally includes calculating an estimate of at least the first feedback action based on the calculated estimate of the user state by using a model including correspondence or correlation data between one or more user states and one or more feedback actions, as described elsewhere herein.

[0373] In this case, the estimation step optionally includes identifying one or more feedback actions associated with a set including a threshold number of user factors indicative of a first state of the user satisfying respective predetermined criteria, as described elsewhere herein.

[0374] Similarly, in this case, the estimation step optionally includes identifying one or more feedback actions associated with leaf nodes of one or more decision trees including branches corresponding to multiple sets or multiple subsets of user factors indicative of a first state of the user satisfying respective predetermined criteria, as described elsewhere herein.

[0375] Similarly, in this case, the model optionally embodies correspondence or correlation data in one or more selected from the list consisting of one or more lookup tables that associate one or more user states with one or more feedback actions, and one or more heuristics that use one or more user states as inputs to identify one or more feedback actions as outputs, as described elsewhere herein.

[0376] Similarly, in this case, the model optionally embodies the correspondence or correlation data in one or more machine learning models that use one or more user states as inputs to identify one or more feedback actions as outputs, as described elsewhere herein.

[0377] Here, the machine learning model is optionally trained based on input including user states formatted as one or more selected from the list of single representative values, representative categories, and multivariate representations, as described elsewhere herein.

[0378] Also herein, the machine learning model is optionally trained based on inputs including self-reported estimates of the user's state obtained from the user, as described elsewhere herein.

[0379] Also herein, the machine learning model is optionally trained based on inputs that also include at least a subset of user factors, as described elsewhere herein.

[0380] Similarly, in this case, the estimation step optionally includes identifying one or more suggested feedback actions for one or more selected from the list consisting of behavioral feedback actions affecting at least a first behavior of the user, drug feedback actions affecting consumption of the active ingredient by the user, and non-consummatory feedback actions affecting one or more non-consummatory actions of the delivery ecosystem, as described elsewhere herein.

[0381] In one example of this summary embodiment, the one-stage correspondence or correlation includes a first correlation between one or more user factors indicative of an acquired user state and at least a first feedback action, as described elsewhere herein.

[0382] In this example, the estimation step optionally includes identifying one or more feedback actions associated with a set including a threshold number of user factors indicative of a first state of the user that satisfies respective predetermined criteria, as described elsewhere herein.

[0383] Similarly, in this example, the estimation step optionally includes identifying one or more feedback actions associated with leaves of one or more decision trees, each of which includes branches corresponding to multiple sets or multiple subsets of user factors indicative of a first state of the user that satisfies each predetermined criterion, as described elsewhere herein.

[0384] Similarly, in this example, the estimation step optionally includes calculating an estimate of at least the first feedback action based on the acquired one or more user factors by using a model including correspondence or correlation data between the one or more feedback actions and the acquired one or more user factors, as described elsewhere in this specification.

[0385] In this case, the model optionally embodies the correspondence or correlation data in one or more selected from the list of: one or more lookup tables that associate one or more user factors with one or more feedback actions, and one or more heuristics that use one or more user factors as inputs to identify one or more feedback actions as outputs, as described elsewhere herein.

[0386] Similarly, in this case, the model optionally embodies the correspondence or correlation data in one or more machine learning models that use one or more user factors as inputs to identify one or more feedback actions as outputs, as described elsewhere herein.

[0387] Here, the machine learning model is optionally trained based on inputs including one or more selected from the list consisting of one or more individual values ​​each based on one or more user factors, one or more combined values ​​each based on two or more user factors, and one or more values ​​each based on user factors from a single data class, as described elsewhere herein.

[0388] Also herein, the machine learning model is optionally trained based on inputs including self-reported estimates of the user's state obtained from the user, as described elsewhere herein.

[0389] Also herein, the machine learning model is optionally trained based on target output feedback actions including one or more selected from the list consisting of behavioral feedback actions affecting at least a first behavior of the user, drug feedback actions affecting consumption of the active ingredient by the user, and non-consummatory feedback actions affecting one or more non-consummatory actions of the delivery ecosystem, as described elsewhere in this specification.

[0390] In one example of this general embodiment, the one or more user factors are each based on one selected from the list consisting of at least a first physical characteristic associated with at least a first user inhalation action, at least a first physical characteristic associated with a user behavior other than inhalation, at least a first physical characteristic associated with a user physiology other than inhalation related, and at least a first aspect of the user's situation separate from handling or operation of the delivery device, as described elsewhere herein.

[0391] In one example of this general embodiment, the one or more user factors are each associated with at least one class selected from the list consisting of historical data providing background information about the user, neurological data about the user, physiological data about the user, contextual data about the user, and environmental data about the user, as described elsewhere herein.

[0392] In one example of this summary embodiment, the user feedback method includes a feedback step that includes selecting, for at least a first device in the delivery ecosystem, at least a first feedback action identified by the estimation step, as described elsewhere herein.

[0393] In this example, the feedback step optionally includes modifying one or more operations of at least a first device in the delivery ecosystem in accordance with the selected feedback action, as described elsewhere herein.

[0394] In this case, the device in the delivery ecosystem whose operation or operations are modified is optionally a delivery device (10), as described elsewhere herein.

[0395] In one instance of this summary embodiment, the selected feedback action includes prompting the user to provide feedback regarding the estimated user state to the user feedback system, as described elsewhere herein.

[0396] In one example of this summary embodiment, the selected feedback action also includes prompting the user to provide feedback regarding the selected feedback action to the user feedback system, as described elsewhere herein.

[0397] In one example of this summary embodiment, the feedback step includes selecting at least a first identified feedback action in response to a current availability of each device in the delivery ecosystem for execution of the feedback action, as described elsewhere herein.

[0398] In one implementation of this general embodiment, the feedback step includes automatically executing at least a first identified feedback action, as described elsewhere herein.

[0399] In one example of this summary embodiment, the feedback step includes prompting the user for consent to have at least some of the identified at least first feedback actions performed, as described elsewhere herein, and, if consent is determined, performing only at least some of the identified at least first feedback actions. Of course, such consent may itself be indicative of the welcomeness or appropriateness of the feedback actions to the user at that time, and thus may be used as input for training a machine learning system.

[0400] In one example of this summary embodiment, the delivery ecosystem comprises one or more selected from the list consisting of one or more delivery devices (10), one or more mobile terminals (100), one or more wearable devices (400), and one or more docking units (200) for the delivery devices, as described elsewhere herein.

[0401] In one example of this general embodiment, at least a portion of the functionality of one or more of the acquisition, estimation, and feedback steps are provided by a remote server (1000), as described elsewhere herein.

[0402] In one example of this general embodiment, at least a portion of the functionality of one or more of the acquisition, estimation, and feedback steps are provided by one or more processors located within one or more devices (10, 100, 200, 300, 400) of the delivery ecosystem (1), as described elsewhere herein.

[0403] It will be appreciated that the methods described above may be implemented on conventional hardware suitably adapted to be applied by the inclusion or substitution of software instructions or dedicated hardware, examples of such hardware having been described herein, for example, in connection with the server 1000 and devices of the delivery ecosystem 1.

[0404] Thus, the adaptation required for existing parts of the conventional equivalent device may be implemented in the form of a computer program product including processor executable instructions stored on a non-transitory machine-readable medium such as a floppy disk, optical disk, hard disk, semiconductor disk, PROM, RAM, flash memory, or any combination thereof, or other storage medium, or may be realized in hardware as an ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), or other configurable circuit suitable for use in adapting the conventional equivalent device. Alternatively, such computer program may be transmitted via data signals over a network such as an Ethernet, a wireless network, the Internet, or any combination thereof, or other network. The present disclosure includes the following embodiments. (Embodiment 1) 1. A user feedback system for a user of a delivery device in a delivery ecosystem, comprising: an acquisition processor configured to acquire one or more user factors indicative of a state of the user; an inference processor configured to identify a one-step or two-step correspondence between one or more user factors indicative of the obtained user state and at least a first feedback action, the feedback action being expected to change the user state as at least in part indicated by the one or more user factors; A user feedback system comprising: (Embodiment 2) The two-step response is as follows: a first correspondence between one or more user factors indicative of the obtained user state and at least a first state of the user; a second correspondence between at least a first state of the user and at least a first feedback action; and 2. The user feedback system of embodiment 1, comprising: (Embodiment 3) 3. The user feedback system of embodiment 2, wherein the estimation processor is operable to calculate an estimate of at least a first state of the user based on the acquired one or more user factors by using a model including correspondence data between one or more user factors and one or more user states. (Embodiment 4) The model is i. one or more lookup tables that associate one or more user factors with one or more user states; ii. one or more heuristics that use one or more user factors as inputs to generate one or more user states as outputs; 4. The user feedback system of embodiment 3, wherein the corresponding data is embodied in one or more selected from the list consisting of: (Embodiment 5) A user feedback system according to any one of embodiments 2 to 4, wherein the estimation processor is configured to calculate an estimate of the first state of the user based on whether a threshold number of user factors indicating at least a first state of the user satisfy each of predetermined criteria. (Embodiment 6) A user feedback system as described in embodiment 5, wherein the number of user factors indicating the first state of the user that meet each predetermined criterion indicates the likelihood that the user is in the first state. (Embodiment 7) A user feedback system as described in embodiment 5 or 6, wherein a subset of user factors indicating the first state of the user is selected, and the threshold number is a part of the subset. (Embodiment 8) 8. The user feedback system of embodiment 7, wherein the subset of user factors is selected based at least in part on their current availability. (Embodiment 9) A user feedback system as described in any one of embodiments 2 to 8, wherein the estimation processor is configured to calculate an estimate of the first state of the user by using one or more decision trees including branches corresponding to a plurality of sets or a plurality of subsets of user factors indicative of at least a first state of the user satisfying respective predetermined criteria, and a leaf of the or each decision tree corresponds to an indication of whether the user is in the at least first state. (Embodiment 10) 10. The user feedback system of claim 9, wherein the length of each branch of the decision tree, with respect to a leaf indicating that the user is in the first state, indicates the likelihood that the user is in the first state. (Embodiment 11) A user feedback system as described in embodiment 3 or 4, wherein the model embodies the correlation data in one or more machine learning models that use one or more user factors as inputs to generate one or more user states as outputs. (Embodiment 12) The machine learning model, i. a self-reported estimate of a user state obtained from said user; ii. an estimate of the user state obtained by a separate analysis of the user factors; and iii. an estimate of a user state derived from the measured effect of the identified at least a first feedback action; and iv. an estimate of a user state derived from self-reported estimates obtained from the user regarding the effect of the identified at least a first feedback action; and 12. The user feedback system of claim 11, wherein the user feedback system is trained based on a target output user state including one or more selected from the list consisting of: (Embodiment 13) A user feedback system as described in any one of embodiments 2 to 12, wherein the estimation processor is operable to identify at least a first feedback action based on the calculated estimate of the user state by using a model including correspondence data between one or more user states and one or more feedback actions. (Embodiment 14) The model is i. one or more lookup tables that associate one or more user states with one or more feedback actions; ii. one or more heuristics that use one or more user states as inputs to identify one or more feedback actions as outputs; 14. The user feedback system of embodiment 13, wherein the corresponding data is embodied in one or more selected from the list consisting of: (Embodiment 15) A user feedback system as described in embodiment 13 or 14, wherein the estimation processor is configured to identify one or more feedback actions associated with a set including a threshold number of user factors indicative of the first state of the user that satisfy each predetermined criterion. (Embodiment 16) A user feedback system described in any one of embodiments 13 to 15, wherein the estimation processor is configured to identify one or more feedback actions associated with leaf nodes of one or more decision trees including branches corresponding to multiple sets or multiple subsets of user factors indicating the first state of the user that satisfy each predetermined criterion. (Embodiment 17) A user feedback system as described in embodiment 13 or 14, wherein the model embodies the corresponding data in one or more machine learning models that use one or more user states as inputs to identify one or more feedback actions as outputs. (Embodiment 18) The estimation processor: i. a behavior feedback action that affects at least a first behavior of the user; ii. a drug feedback action that affects consumption of an active ingredient by said user; and iii. a non-consumable feedback action that affects one or more non-consumable actions of the delivery ecosystem; and 18. A user feedback system as described in any one of embodiments 13 to 17, operable to identify one or more suggested feedback actions relating to one or more selected from the list consisting of: (Embodiment 19) 2. The user feedback system of embodiment 1, wherein the one-stage correspondence includes a first correspondence between one or more user factors indicative of the acquired user state and at least a first feedback action. (Embodiment 20) A user feedback system as described in embodiment 19, wherein the estimation processor is operable to identify at least a first feedback action based on the acquired one or more user factors by using a model including correspondence data between one or more feedback actions and the acquired one or more user factors. (Embodiment 21) The model is i. one or more lookup tables that associate one or more user factors with one or more feedback actions; ii. one or more heuristics that use one or more user factors as inputs to identify one or more feedback actions as outputs; 21. The user feedback system of embodiment 20, wherein the correlation data is embodied in one or more selected from the list consisting of: (Embodiment 22) A user feedback system as described in any one of embodiments 19 to 21, wherein the estimation processor is configured to identify one or more feedback actions associated with a set including user factors among a threshold number of user factors that indicate the first state of the user that satisfy each predetermined criterion. (Embodiment 23) A user feedback system as described in any one of embodiments 19 to 22, wherein the estimation processor is configured to identify one or more feedback actions associated with leaves of one or more decision trees, each of which includes branches corresponding to multiple sets or multiple subsets of user factors indicating the first state of the user that satisfy each predetermined criterion. (Embodiment 24) A user feedback system as described in embodiment 18 or 19, wherein the model embodies the corresponding data in one or more machine learning models that use one or more user factors as inputs to identify one or more feedback actions as outputs. (Embodiment 25) Each user factor of the one or more user factors: i. at least a first physical characteristic associated with at least a first user inhalation action; ii. at least a first physical characteristic associated with a user behavior other than inhalation; and iii. at least a first physical characteristic associated with a non-aspiration-related physiology of a user; and iv. at least a first aspect of the user's context separate from the handling or operation of the delivery device; and The user feedback system according to any one of embodiments 1 to 24, based on one selected from the list consisting of: (Embodiment 26) The one or more user factors each include: i. historical data providing background information about the user; ii. neurological data about the user; and iii. physiological data relating to the user; and iv. contextual data about the user; and v. environmental data about the user; 26. A user feedback system according to any one of embodiments 1 to 25, relating to at least one class selected from the list consisting of: (Embodiment 27) A user feedback system described in any one of embodiments 1 to 26, comprising a feedback processor configured to select at least a first feedback action identified by the estimation processor for at least a first device in the delivery ecosystem. (Embodiment 28) 28. A user feedback system as described in embodiment 27, wherein the feedback processor is configured to modify one or more operations of at least a first device in the delivery ecosystem in accordance with the or each selected feedback action. (Embodiment 29) 29. The user feedback system of embodiment 28, wherein the device in the delivery ecosystem whose one or more operations are modified is the delivery device. (Embodiment 30) 30. A user feedback system according to any one of embodiments 1 to 29, wherein the selected feedback action also includes prompting the user to provide feedback regarding the selected feedback action to the user feedback system. (Embodiment 31) A user feedback system as described in any one of embodiments 1 to 30, wherein the feedback processor (1030) is configured to select the identified at least a first feedback action in response to a current availability of each device in the delivery ecosystem for execution of a feedback action. (Embodiment 32) 32. A user feedback system according to any one of embodiments 1 to 31, wherein the feedback processor is configured to automatically execute the identified at least first feedback action. (Embodiment 33) A user feedback system described in any one of embodiments 1 to 32, wherein the feedback processor is configured to prompt a user for consent to having at least a portion of the identified at least first feedback action executed, and if consent is determined, to have only the at least a portion of the identified at least first feedback action executed. (Embodiment 34) The delivery ecosystem comprises: i. one or more delivery devices; ii. one or more mobile terminals; iii. one or more wearable devices; iv. one or more docking units for the or each delivery device; and A user feedback system according to any one of embodiments 1 to 33, comprising one or more selected from the list consisting of: (Embodiment 35) A user feedback system according to any one of embodiments 1 to 34, wherein at least a portion of the functionality of one or more of the acquisition processor, the estimation processor, and the feedback processor is provided by a remote server. (Embodiment 36) A user feedback system described in any one of embodiments 1 to 35, wherein at least a portion of the functions of one or more of the acquisition processor, estimation processor, and feedback processor are provided by one or more processors located within one or more devices of the delivery ecosystem. (Embodiment 37) 1. A method of user feedback to a user of a delivery device (10) within a delivery ecosystem, comprising: acquiring one or more user factors indicative of a state of the user; - an inference step of identifying a one- or two-step correspondence between one or more user factors indicative of the obtained user state and at least a first feedback action, the feedback action being expected to change the user state as at least partially indicated by the one or more user factors; A user feedback method comprising: (Embodiment 38) The two-step response is as follows: a first correspondence between one or more user factors indicative of the obtained user state and at least a first state of the user; a second correspondence between at least a first state of the user and at least a first feedback action; and 38. The user feedback method of embodiment 37, comprising: (Embodiment 39) A user feedback method as described in embodiment 38, wherein the estimation step includes calculating an estimate of at least a first state of the user based on the acquired one or more user factors by using a model including correspondence data between one or more user factors and one or more user states. (Embodiment 40) A user feedback method as described in embodiment 38 or 39, wherein the estimation step includes calculating an estimate of the first state of the user based on whether a threshold number of user factors indicative of at least a first state of the user satisfy each of predetermined criteria. (Embodiment 41) A user feedback method as described in embodiment 38 or 39, wherein the estimation step includes calculating an estimate of the first state of the user by using one or more decision trees including branches corresponding to multiple sets or multiple subsets of user factors indicative of at least a first state of the user satisfying respective predetermined criteria, and a leaf of the or each decision tree corresponds to an indication of whether the user is in the at least first state. (Embodiment 42) A user feedback method according to any one of embodiments 37 to 41, wherein the estimation step includes identifying at least a first feedback action based on the calculated estimated value of the user state by using a model including correspondence data between one or more user states and one or more feedback actions. (Embodiment 43) A user feedback method as described in embodiment 37, wherein the one-stage correspondence includes a first correspondence between one or more user factors indicative of the acquired user state and at least a first feedback action. (Embodiment 44) A user feedback method as described in embodiment 43, wherein the estimation step includes calculating an estimate of at least a first feedback action based on the acquired one or more user factors by using a model including corresponding correlation data between one or more feedback actions and the acquired one or more user factors. (Embodiment 45) A user feedback method as described in embodiment 43 or 44, wherein the estimation step includes identifying one or more feedback actions associated with a set including user factors among a threshold number of user factors that indicate the first state of the user that satisfy each predetermined criterion. (Embodiment 46) A user feedback method described in any one of embodiments 43 to 46, wherein the estimation step includes identifying one or more feedback actions associated with leaves of one or more decision trees each including branches corresponding to multiple sets or multiple subsets of user factors indicating the first state of the user that satisfy each predetermined criterion. (Embodiment 47) A user feedback method according to any one of embodiments 43 to 46, wherein the correlation data is embodied in one or more machine learning models, the models using one or more user factors as inputs to identify one or more feedback actions as outputs. (Embodiment 48) A computer program comprising computer-executable instructions configured to cause a computer system to perform the method according to any one of embodiments 37 to 47. (Embodiment 49) A computer program product comprising the computer program of embodiment 48 stored on a non-transitory machine-readable medium.

Claims

1. 1. A user feedback system for a user of a delivery device in a delivery ecosystem, comprising: an acquisition processor configured to acquire one or more user factors indicative of a state of the user; an inference processor configured to identify a two-level correspondence between one or more user factors indicative of the obtained user state and at least a first feedback action, the two-level correspondence comprising: a first correspondence between one or more user factors indicative of the obtained user state and at least a first state of the user; a second correspondence between at least a first state of the user and at least a first feedback action; and Including, an estimation processor, wherein the feedback action is expected to change a state of the user as indicated at least in part by the one or more user factors; A user feedback system comprising:

2. 2. The user feedback system of claim 1, wherein the estimation processor is operable to calculate an estimate of at least a first state of the user based on the acquired one or more user factors by using a model comprising correspondence data between one or more user factors and one or more user states.

3. The model is i. one or more lookup tables that associate one or more user factors with one or more user states; ii. one or more heuristics that use one or more user factors as inputs to generate one or more user states as outputs; The user feedback system of claim 2 , further comprising: a) a user interface for enabling a user to communicate with a user through a network;

4. 4. The user feedback system of claim 1, wherein the estimation processor is configured to calculate an estimate of the first state of the user based on whether a threshold number of user factors indicative of at least a first state of the user satisfy respective predetermined criteria.

5. The user feedback system of claim 4 , wherein the number of user factors indicative of the first state of the user that satisfy each predetermined criterion indicates the likelihood that the user is in the first state.

6. 6. The user feedback system of claim 4 or 5, wherein a subset of user factors indicative of the first state of the user is selected, and the threshold number is a part of the subset.

7. The user feedback system of claim 6 , wherein the subset of user factors is selected based at least in part on their current availability.

8. 8. The user feedback system of claim 1, wherein the estimation processor is configured to calculate the estimate of the first state of the user by using one or more decision trees comprising branches corresponding to a plurality of sets or subsets of user factors indicative of at least a first state of the user satisfying respective predetermined criteria, and wherein leaves of the one or more decision trees correspond to an indication of whether the user is in the at least first state.

9. 9. The user feedback system of claim 8, wherein, for a leaf indicating that the user is in the first state, the length of each branch of the decision tree indicates the likelihood that the user is in the first state.

10. 4. The user feedback system of claim 2 or 3, wherein the model embodies the correlation data in one or more machine learning models that use one or more user factors as inputs to generate one or more user states as outputs.

11. The machine learning model, i. a self-reported estimate of a user's state obtained from said user; ii. an estimate of the user state obtained by a separate analysis of the user factors; iii. an estimate of a user state derived from a measured effect of the identified at least a first feedback action; iv. an estimate of a user state derived from self-reported estimates obtained from the user regarding the effect of the identified at least a first feedback action; 11. The user feedback system of claim 10, wherein the user feedback system is trained based on target output user states including one or more selected from the list consisting of:

12. 12. A user feedback system according to claim 2 or any one of claims 3 to 11 when directly or indirectly dependent on claim 2, wherein the estimation processor is operable to identify at least a first feedback action based on the calculated estimate of a user state by using a model comprising correspondence data between one or more user states and one or more feedback actions.

13. The model is i. one or more lookup tables that associate one or more user states with one or more feedback actions; ii. one or more heuristics that use one or more user states as inputs to identify one or more feedback actions as outputs; The user feedback system of claim 12 , further comprising: a) a step of: embodiing said corresponding data in one or more selected from the list consisting of:

14. 14. The user feedback system of claim 12 or 13, wherein the inference processor is configured to identify one or more feedback actions associated with a set including a threshold number of user factors indicative of the first state of the user satisfying respective predetermined criteria.

15. 15. The user feedback system of claim 12, wherein the inference processor is configured to identify one or more feedback actions associated with leaf nodes of one or more decision trees including branches corresponding to multiple sets or multiple subsets of user factors indicative of the first state of the user satisfying respective predetermined criteria.

16. 14. The user feedback system of claim 12 or 13, wherein the model embodies the corresponding data in one or more machine learning models that use one or more user states as inputs to identify one or more feedback actions as outputs.

17. The estimation processor: i. a behavior feedback action that affects at least a first behavior of the user; ii. a drug feedback action that affects the consumption of an active ingredient by the user; iii. A non-consumable feedback action that affects one or more non-consumable actions of the delivery ecosystem; A user feedback system according to any one of claims 12 to 16, operable to identify one or more suggested feedback actions relating to one or more selected from the list consisting of:

18. Each user factor of the one or more user factors: i. at least a first physical characteristic associated with at least a first user inhalation action; ii. at least a first physical characteristic associated with a user behavior other than inhalation; iii. at least a first physical characteristic associated with a non-aspiration-related physiology of a user; iv. at least a first aspect of the user's context separate from the handling or operation of the delivery device; The user feedback system according to any one of claims 1 to 17, based on one selected from the list consisting of:

19. The one or more user factors each include: i. historical data providing background information about the user; ii. neurological data about the user; iii. physiological data relating to the user; iv. contextual data about the user; v. environmental data regarding said user; A user feedback system according to any one of claims 1 to 18, relating to at least one class selected from the list consisting of:

20. 20. The user feedback system of any one of claims 1 to 19, comprising a feedback processor configured to select at least a first feedback action identified by the inference processor for at least a first device in the delivery ecosystem.

21. 21. The user feedback system of claim 20, wherein the feedback processor is configured to modify one or more operations of at least a first device in the delivery ecosystem according to the at least a first selected feedback action.

22. 22. The user feedback system of claim 21, wherein the device in the delivery ecosystem for which one or more operations are modified is the delivery device.

23. 23. The user feedback system of claim 21 or 22, wherein the feedback processor is configured to select the identified at least a first feedback action in response to a current availability of each device in the delivery ecosystem for execution of a feedback action.

24. A user feedback system according to any one of claims 21 to 23, wherein the feedback processor is configured to automatically execute the identified at least a first feedback action.

25. 25. A user feedback system according to any one of claims 21 to 24, wherein the feedback processor is configured to prompt a user for consent to have at least a portion of the identified at least first feedback actions executed, and if consent is determined, to have only the at least a portion of the identified at least first feedback actions executed.

26. A user feedback system as described in any one of claims 1 to 25, wherein the selected feedback action also includes prompting the user to provide feedback regarding the selected feedback action to the user feedback system.

27. The delivery ecosystem comprises: i. one or more delivery devices; ii. one or more mobile terminals; iii. one or more wearable devices; iv. one or more docking units for said one or more delivery devices; A user feedback system according to any preceding claim, comprising one or more selected from the list consisting of:

28. A user feedback system according to any preceding claim, wherein at least some of the functionality of one or more of the acquisition processor and the estimation processor is provided by a remote server.

29. 29. A user feedback system according to any one of claims 1 to 28, wherein at least some of the functions of one or more of the acquisition processor and the estimation processor are provided by one or more processors located within one or more devices of the delivery ecosystem.

30. A method of user feedback to a user of a delivery device (10) in a delivery ecosystem, comprising: acquiring one or more user factors indicative of a state of the user; an estimation step of identifying a two-level correspondence between one or more user factors indicative of the acquired user state and at least a first feedback action, the two-level correspondence comprising: a first correspondence between one or more user factors indicative of the obtained user state and at least a first state of the user; a second correspondence between at least a first state of the user and at least a first feedback action; and Including, an estimating step, in which the feedback action is expected to change a state of the user as at least in part indicated by the one or more user factors; A user feedback method comprising:

31. 31. The user feedback method of claim 30, wherein the estimating step comprises calculating an estimate of at least a first state of the user based on the obtained one or more user factors by using a model comprising correspondence data between one or more user factors and one or more user states.

32. 32. The user feedback method according to claim 30 or 31, wherein the estimating step comprises calculating an estimate of the first state of the user based on whether a threshold number of user factors indicative of at least a first state of the user satisfy respective predetermined criteria.

33. 32. The user feedback method of claim 30 or 31, wherein the estimating step comprises calculating an estimate of the first state of the user by using one or more decision trees including branches corresponding to a plurality of sets or subsets of user factors indicative of at least a first state of the user satisfying respective predetermined criteria, and leaves of the one or more decision trees corresponding to an indication of whether the user is in the at least first state.

34. 34. A user feedback method according to claim 31 or claim 32 or 33 when dependent on claim 31, wherein the estimating step comprises identifying at least a first feedback action based on the calculated estimate of the user state by using a model comprising correspondence data between one or more user states and one or more feedback actions.

35. A computer program comprising computer executable instructions arranged to cause a computer system to carry out a method according to any one of claims 30 to 34.

36. 36. A computer program product comprising the computer program of claim 35 stored on a non-transitory machine readable medium.

Images