DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Status of Claims
Claims 1-20 are pending of which claims 1 and 13-16 are in independent form.
Claims 1-20 are rejected under 35 U.S.C. 103.
Response to Arguments
Applicant's arguments filed 7/23/2026 have been fully considered but they are not persuasive.
Regarding 35 USC 101
Applicant’s arguments, see “Remarks”, filed 7/23/2026, with respect to 35 USC 101 (Abstract Idea) have been fully considered and are persuasive. The 35 USC 101 rejection of claims 1-20 has been withdrawn.
Regarding 35 USC 103
Applicant’s arguments (pages 16-21 of the “Remarks”) regarding the proposed combination of Sajda, Myrden and Alcaide and alleged hindsight reconstruction have been considered but are not persuasive.
Contrary to Applicant’s assertions, Sajda teaches classifying physiological responses evoked by environmental objects to identify objects of interest and associating resulting classification with computer vision feature (see ¶ [0035]-[0036], [0040]-[0042], Fig. 3). Sajda further teaches correlation physiological responses with environment events to infer cognitive states and adaptive AI agent behavior based on physiological and contextual information (see ¶ [0047], [0066]). Thus, Sajda already provides the underlaying relationship between physiological classification, contextual association, and ML adaptation.
Myrden is introduced specifically for EEG acquisition and brain signal classification (see ¶ [0117]-[0123], [0148]-[0150]), while Alcaide provides physical scene sensing and identification of environment incentives (see ¶ [0031]-[0033]). It would have been obvious to incorporate these complementary techniques into Sajda to facilitate identifying environmental objects associated with physiological responses and using such information to adapt AI operations, with a reasonable exception of success given the compatible sensing and processing functions. Accordingly, the proposed combination is supported by the references’ teachings rather than hindsight reconstruction, and therefore the rejection under 35 USC 103 is maintained.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Sajda; Paul et al. (US 20190101985 A1) [Sajda] in view of MYRDEN; Andrew et al. (US 20190212816 A1) [Myrden] in view of ALCAIDE; Ramses et al. (US 20200192478 A1) [Alcaide].
Regarding claim 1, Sajda discloses, a computer-implemented method for adapting a machine-learning algorithm executed by an information processing device for processing data occurring in a real-life context (FIG. 2 is a block diagram illustrating a system level diagram of the disclosed hBCI system 200. The hBCI system 200 can include one or more sensory input devices 202 as shown in FIG. 1, a machine learning module 210 that can process input signals from sensory input device(s) 202 to reinforce driving behavior using a deep reinforcement network, and an environment module 204 that can generate a virtual environment in which the AI agent can drive a simulated autonomous vehicle using the determined reinforced driving behavior ¶ [0033], [0034]-[0036], also see Figs. 2 and 3), said machine-learning algorithm processing:
multi-modal operational data representing contextual aspects occurring in said real-life context, [said multi-modal operational data provided by at least one context sensor] and comprising at least one of sensed, perceived, obtained and otherwise identified physical data, virtual data and environmental information continuously originating from said real- life context (The disclosed systems and methods provide a hybrid brain-computer-interface (hBCI) that can detect an individual's implicit reinforcement signals (e.g., level of interest, arousal, emotional reactivity, cognitive fatigue, cognitive state, or the like) in response objects, events, and/or actions in an environment by generating implicit reinforcement signals for improving an AI agent controlling, by way of example only, a simulated autonomous vehicle. As utilized herein, an “environment” can include a real-world environment, a virtual reality environment, and/or an augmented reality environment. Additionally, the disclosed subject matter provides systems and methods for using a brain-artificial intelligence (AI) interface to use neurophysiological signals to help deep reinforcement learning based AI systems adapt to human expectations as well as increase task performance. The disclosed subject matter can provide access to the cognitive state of humans, such as through real-time electroencephalogram (EEG) based neuroimaging, and can be used to understand implicit non-verbal cues in real-time. Furthermore, the systems and methods disclosed can use neurophysiological signals to inform deep reinforcement learning based AI systems to enhance user comfort and/or trust in automation ¶ [0020]-[0023], [0027], [0029], [0031]-[0033], [0045], [0049]);
neural brain activity data comprising neurophysiological response data relating to implicit human neurophysiological participation of a human participant participating with said real-life context (The disclosed systems and methods provide a hybrid brain-computer-interface (hBCI) that can detect an individual's implicit reinforcement signals (e.g., level of interest, arousal, emotional reactivity, cognitive fatigue, cognitive state, or the like) in response objects, events, and/or actions in an environment by generating implicit reinforcement signals for improving an AI agent controlling, by way of example only, a simulated autonomous vehicle. As utilized herein, an “environment” can include a real-world environment, a virtual reality environment, and/or an augmented reality environment. Additionally, the disclosed subject matter provides systems and methods for using a brain-artificial intelligence (AI) interface to use neurophysiological signals to help deep reinforcement learning based AI systems adapt to human expectations as well as increase task performance. The disclosed subject matter can provide access to the cognitive state of humans, such as through real-time electroencephalogram (EEG) based neuroimaging, and can be used to understand implicit non-verbal cues in real-time. Furthermore, the systems and methods disclosed can use neurophysiological signals to inform deep reinforcement learning based AI systems to enhance user comfort and/or trust in automation ¶ [0020]-[0022], [0029], [0040], [0051]-[0052], [0056], also see Figs. 3 and 6), and engagement data referring to [what aspect or aspects of said real-life context] said human participant is momentarily implicitly involved with (FIG. 3 is a block diagram illustrating a system level diagram of the disclosed machine learning module 300 in accordance with the present disclosure. Machine learning module 300 can correspond to machine learning module 210 of FIG. 2. The machine learning module 300 can include a hBCI submodule, a TAG module, and a reinforcement learning module. The hBCI module can use a hybrid classifier to fuse and decode physiological signals from the subject and identify target or non-target labels from a subset of the object database. As shown in FIG. 3, the TAG module can use those hBCI object labels, generated by the hBCI module, as features along with a computer (CV) system to identify targets and non-targets from a large dataset of objects. The reinforcement learning module can populate a virtual environment with objects from the large dataset and use those labels generated by the TAG module to elicit reinforcement to the AI agent in order to train the AI agent to keep targets in view for longer periods of time ¶ [0035]-[0035], [0051], [0052], also see Figs. 3 and 6),
wherein said neural brain activity data is provided by a passive brain-computer interface operating by: [continuously acquiring, from at least one human head-based sensor, brain activity signals] of said human participant participating with said real-life context (The disclosed systems and methods provide a hybrid brain-computer-interface (hBCI) that can detect an individual's implicit reinforcement signals (e.g., level of interest, arousal, emotional reactivity, cognitive fatigue, cognitive state, or the like) in response objects, events, and/or actions in an environment by generating implicit reinforcement signals for improving an AI agent controlling, by way of example only, a simulated autonomous vehicle. As utilized herein, an “environment” can include a real-world environment, a virtual reality environment, and/or an augmented reality environment. Additionally, the disclosed subject matter provides systems and methods for using a brain-artificial intelligence (AI) interface to use neurophysiological signals to help deep reinforcement learning based AI systems adapt to human expectations as well as increase task performance. The disclosed subject matter can provide access to the cognitive state of humans, such as through real-time electroencephalogram (EEG) based neuroimaging, and can be used to understand implicit non-verbal cues in real-time. Furthermore, the systems and methods disclosed can use neurophysiological signals to inform deep reinforcement learning based AI systems to enhance user comfort and/or trust in automation ¶ [0020]-[0022], [0029], [0031], [0034]-[0035], see Figs. 1-3); and
classifying, by at least one classifier, [from said brain activity signals] neurophysiological response data relating to implicit human neurophysiological participation of said human participant with said real-life context (The hBCI module can use a hybrid classifier to fuse and decode physiological signals from the subject and identify target or non-target labels from a subset of the object database ¶ [0035], [0052], [0059], [0062], [0063]. Also see ¶ [0040], [0051], Figs. 3 and 6), and
classifying, by at least one classifier, [from said brain activity signals engagement data] referring to what aspect or aspects of said real-life context said human participant is momentarily implicitly involved with (The hBCI module can use a hybrid classifier to fuse and decode physiological signals from the subject and identify target or non-target labels from a subset of the object database ¶ [0035], [0052], [0059], [0062], [0063]. Also see ¶ [0040], [0051], Figs. 3 and 6);
acquiring, by said passive brain-computer interface, [from said at least one human head-based sensor], brain activity signals of a human participant participating with said real-life context (Sajda: acquiring brain activity signals through EEG physiological monitoring ¶ [0020], [0029], [0031]; also see Figs. 1 and 2);
classifying, by at least one classifier, neurophysiological response data relating to implicit human neurophysiological participation of said human participant, [from said brain activity signals] (Sajda: The hBCI module can use a hybrid classifier to fuse and decode physiological signals from the subject and identify target or non-target labels from a subset of the object database. … classifying naturally evoked neural and physiological responses associated with environmental objects ¶ [0035], [0040], [0051]-[0052], [0059], [0062], [0063]. Also see Figs. 3 and 6);
classifying, by at least one classifier, engagement data referring to what aspect or aspects of said real-life context said human participant is momentarily implicitly involved with, [from said brain activity signals] (Sajda: The hBCI module can use a hybrid classifier to fuse and decode physiological signals from the subject and identify target or non-target labels from a subset of the object database. … classifying naturally evoked neural and physiological responses associated with object interests ¶ [0035], [0040], [0051]-[0052], [0059], [0062], [0063]. Also see Figs. 3 and 6);
acquiring, by said information processing device, [from said at least one context sensor], multi-modal operational data occurring in said real-life context (Sajda: processing environmental perception information, computer vision features and operational vehicle state information ¶ [0031]-[0035], [0040], [0045], [0046]; In certain example embodiments, methods for detecting reinforcement signals of a user or an environment through at least one sensor are disclosed. The methods can include positioning the user in the environment including one or more objects, events, or actions for the user to sense, and collecting sensory information for the user from the at least one sensor ¶ [0006]-[0007], [0021], [0070]; also see Figs. 2, 3, 5A and 5B);
acquiring, by said information processing device, neurophysiological response data and engagement data from said passive-brain computer interface (Sajda: using an hBCI module to produce physiological classification and object interest information for downstream processing ¶ [0034]-[0036], [0051]-[0052], also see Figs. 2-3, and 6);
linking, by said information processing device, acquired multi-modal operational data and acquired neurophysiological response data [associated with an aspect of said real-life context referred to by said acquired engagement data] (Sajda: FIG. 3 is a block diagram illustrating a system level diagram of the disclosed machine learning module 300 in accordance with the present disclosure. Machine learning module 300 can correspond to machine learning module 210 of FIG. 2. The machine learning module 300 can include a hBCI submodule, a TAG module, and a reinforcement learning module. The hBCI module can use a hybrid classifier to fuse and decode physiological signals from the subject and identify target or non-target labels from a subset of the object database. As shown in FIG. 3, the TAG module can use those hBCI object labels, generated by the hBCI module, as features along with a computer (CV) system to identify targets and non-targets from a large dataset of objects. The reinforcement learning module can populate a virtual environment with objects from the large dataset and use those labels generated by the TAG module to elicit reinforcement to the AI agent in order to train the AI agent to keep targets in view for longer periods of time ¶ [0035]-[0036], [0040], [0047], [0051]-[0051]);
assessing, by said information processing device, from said linked neurophysiological response data, at least one type of implicit neurophysiological response of said human participant (FIG. 4 is a diagram illustrating a process by which the disclosed subject matter can be used to infer the passenger's psychological state. Sensory input devices monitoring the passenger and/or subject can measure and/or sense the subject's reactions. By fusing and/or compositing the various different reactions that are measured from the subject and correlating it with the actions and/or events occurring in the environment that the subject is experiencing, the AI agent of the reinforcement learning module (e.g., reinforcement learning module 210 as shown in FIGS. 2 and 3) can infer a cognitive state and user intent of the subject, hereinafter also referred to as the passenger's psychological state. For example, the AI agent controlling the simulated autonomous vehicle can be informed by the psychological state of the subject being driven in the simulated autonomous vehicle and experiencing the environment. The AI agent can be adjusted based on reinforcement from the inferred psychological state of the subject such that the AI agent's actions are affected by the passenger's psychological state. For example, upon determining which types of objects that the subject is interested in, the AI agent can determine when to slow down, brake, and accelerate the simulated autonomous vehicle. The AI agent can adjust these actions based on the inferred interest of the subject in the objects and/or surroundings in the environment. The subject's neurological and/or ocular signals can stem from naturally occurring reactions of the subject. The disclosed system can collect such information provided passively by the subject as he sits in the seat of the simulated autonomous vehicle experiencing the simulated drive in the environment. The AI agent can be trained to allow the subject to view the type of objects that the subject shows an interest in. By measuring and factoring in the different measured ocular and neurological signals of the subject, the AI agent can modify the behavior of the simulated autonomous vehicle to the individual's preferences to inform the simulated autonomous vehicle about the needs of the subject. By customizing the actions of the simulated autonomous vehicle to the individual subject, the simulated autonomous vehicle driving can be customized to the individual passenger and can result in increased comfort, trust, and user acceptance in the AI agent ¶ [0047], [0029], [0040], [0051]-[0052]; Also see Figs. 3-4, 6), and
adapting, said information processing device, based on said assessed at least one type of implicit neurophysiological response of said human participant and said linked multi-modal operational data, operation of at least one of a control operation and inference operation of said machine-learning algorithm performed in relation to said real-life context (In some embodiments, the AI agent can learn a braking strategy that maintains a safe distance from a lead vehicle, minimizes drive time, and can slow the simulated autonomous vehicle when encountering virtual objects of interest that are identified from hBCI signals of the passenger. Example disclosed method(s) allowing for such slowing down of the simulated autonomous vehicle can provide the simulated passenger approximately at least 20% more time to gaze at virtual objects in the virtual environment they find interesting, compared to objects that they have no interest in (or distractor objects and/or events). The disclosed subject matter provides systems and methods to use an hBCI to provide implicit reinforcement to an AI agent in a manner that incorporates user preferences into the control system ¶ [0028]-[0029], [0035]-[0036], [0043]-[0047], [0056], [0065]-[0068]).
However, Sajda does not explicitly facilitate, said multi-modal operational data provided by at least one context sensor; what aspect or aspects of said real-life context; from said at least one context sensor; associated with an aspect of said real-life context referred to by said acquired engagement data.
Alcaide discloses, said multi-modal operational data provided by at least one context sensor; what aspect or aspects of said real-life context (The cameras can be connected to each other, through a wired or wireless connection. The video based eye-tracker can also include an additional scene camera that captures the uses' field of view ¶ [0033], [0031], [0032], also see Figs. 2A and 2B).
from said at least one context sensor (In some embodiments, the video based eye-tracker 102 relies on tracking the user's pupil and a first-surface corneal reflection (CR) of an illumination source with the use of a head-mounted eye tracking video camera to image the user's eye. The positional difference between these two features can be used to determine the observer's eye-in-head orientation. Some example head mounted eye-tracking devices that can be used as the video based eye-tracker 102, are shown in FIGS. 2A and 2B and are available from SenseMotoric Instruments, Tobii Eye Tracking, and Pupil-labs among other commercial vendors. For example, the illustration in FIG. 2B shows a front and rear view of a video based eye-tracker, with the left and right eye cameras indicated. The cameras can be connected to each other, through a wired or wireless connection. The video based eye-tracker can also include an additional scene camera that captures the uses' field of view ¶ [0033], also see Figs. 2A and 2B);
associated with an aspect of said real-life context referred to by said acquired engagement data (In some embodiments of the hybrid BCI system 100, the user's point of focus can be determined from the video based eye-tracker 102 (i.e., the pointing control feature), and the neural recording headset 104 can be used for collecting neural brain activity data (i.e., the action control feature). The collected neural and eye-tracking data can be communicated to the Brain-Computer Interfacing Device 110 that processes the signals as an ensemble along with data about what stimuli were presented. With the combined information, the Brain-Computer interfacing Device 110 can detect relevant signal features based on statistical models to predict the user's intent. This predicted intent can then be communicated to the user, via the user interface presented through the display 106 for example, and acted upon ¶ [0031]-[0033]; In some embodiments, a processor 620 of the Brain-Computer Interfacing Device 610 can include a Synchronized Event Logger 622 that receives the eye-tracking and neural signal data via an L/O unit 640, and timestamps the signal data to be synchronous with each other. In some embodiments, the Synchronized Event Logger 622 can be configured to perform high speed eye movement classification ¶ [0075]);
It would have been obvious to one ordinary skilled in the art at the time of the present invention to combine the teachings of the cited references because Alcaide's system would have allowed Sajda to facilitate said multi-modal operational data provided by at least one context sensor; what aspect or aspects of said real-life context; from said at least one context sensor; associated with an aspect of said real-life context referred to by said acquired engagement data. The motivation to combine is apparent in the Sajda reference, because there is a need to improve BCI system to address the need for Brain Computer Interfaces that operate with high-speed and accuracy.
However, neither Sajda nor Alcaide explicitly facilitates continuously acquiring, from at least one human head-based sensor, brain activity signals; from said brain activity signals; from said brain activity signals engagement data; from said at least one human head-based sensor.
Myrden discloses, continuously acquiring, from at least one human head-based sensor, brain activity signals (A system for a brain-computer interface (BCI) is provided. The system comprises an output unit configured to trigger a series of mental tasks for the patient, a device having a plurality of electrodes to continuously capture real-time raw electroencephalography (EEG) data from a patient, a server, and a display device to display and update an interface with visual elements based on issued control commands from the server [Abstract], ¶ [0003], [0024], [0035], [0056], [0102], [0103], [0108], [0124], [0125], [0196], Gigs, 6A and 6B);
from said brain activity signals; from said brain activity signals engagement data (Classification device 120 can be used to build a classification model by training on data received from interface application 130 or other entities 150, for example, EEG data collected during a change in mental state of a user. Data collection unit 122 associated with a classification device 120 and BCI platform 110 can receive data, for example, EEG data from a single user via interface application 130. Data collection unit 122 can receive stored data from one or more external systems 150 or interface applications 130, for example, corresponding to other sessions of data collection ¶ [0117]-[00123], [0148], [0149]);
from said at least one human head-based sensor (FIG. 6A illustrates, in a flowchart, an example of a method 600a of determining a mental state from multichannel EEG data, in accordance with some embodiments. The EEG data may be received from electrodes 52 located on a head of a patient, on an EEG cap device or on an in-ear EEG device. The method 600a may be performed by the BCI platform 110, classification device 120, and the interface application 130. The method 600a comprises generating 610 a set of features based upon a frequency domain analysis of the EEG data. Next, a dimensionality of the set of features is reduced 620. Next, the features are classified 630 into mental states. The method 600a will be described in more detail below ¶ [0148]-[0150]; In some embodiments, the mental state of the patient is monitored via passive BCI monitoring in parallel with active BCI monitoring ¶ [0015], [0097]; also see Fig. 1);
It would have been obvious to one ordinary skilled in the art at the time of the present invention to combine the teachings of the cited references because Myrden's system would have allowed Sajda and Alcaide to facilitate continuously acquiring, from at least one human head-based sensor, brain activity signals; from said brain activity signals; from said brain activity signals engagement data; from said at least one human head-based sensor. The motivation to combine is apparent in the Sajda and Alcaide reference, because there is a need to improve brain-computer interfaces data collection and utilization.
Regarding claim 2, the combination of Sajda, Myrden and Alcaide discloses, wherein adapting operation of said machine learning algorithm comprises associating [multi-modal operational data] and neural brain activity data [corresponding to said aspect referred to by said acquired engagement] data (Sajda: FIG. 3 is a block diagram illustrating a system level diagram of the disclosed machine learning module 300 in accordance with the present disclosure. Machine learning module 300 can correspond to machine learning module 210 of FIG. 2. The machine learning module 300 can include a hBCI submodule, a TAG module, and a reinforcement learning module. The hBCI module can use a hybrid classifier to fuse and decode physiological signals from the subject and identify target or non-target labels from a subset of the object database. As shown in FIG. 3, the TAG module can use those hBCI object labels, generated by the hBCI module, as features along with a computer (CV) system to identify targets and non-targets from a large dataset of objects. The reinforcement learning module can populate a virtual environment with objects from the large dataset and use those labels generated by the TAG module to elicit reinforcement to the AI agent in order to train the AI agent to keep targets in view for longer periods of time ¶ [0035]-[0036], [0040]-[0042], [0047], Figs. 3-4).
Alcaide discloses, multi-modal operational data … corresponding to said aspect referred to by said acquired engagement (In some embodiments of the hybrid BCI system 100, the user's point of focus can be determined from the video based eye-tracker 102 (i.e., the pointing control feature), and the neural recording headset 104 can be used for collecting neural brain activity data (i.e., the action control feature). The collected neural and eye-tracking data can be communicated to the Brain-Computer Interfacing Device 110 that processes the signals as an ensemble along with data about what stimuli were presented. With the combined information, the Brain-Computer interfacing Device 110 can detect relevant signal features based on statistical models to predict the user's intent. This predicted intent can then be communicated to the user, via the user interface presented through the display 106 for example, and acted upon ¶ [0031]-[0033], [0075]).
Regarding claim 3, the combination of Sajda, Myrden and Alcaide discloses, wherein adapting operation of said machine learning algorithm is performed based on at least one of: an aspect referred to by said acquired [engagement data in relation to specific operations and interactions occurring in said real-life context], and interaction of said information processing device with specific operations and interactions occurring in said real-life context (Sajda: In some embodiments, the AI agent can learn a braking strategy that maintains a safe distance from a lead vehicle, minimizes drive time, and can slow the simulated autonomous vehicle when encountering virtual objects of interest that are identified from hBCI signals of the passenger. Example disclosed method(s) allowing for such slowing down of the simulated autonomous vehicle can provide the simulated passenger approximately at least 20% more time to gaze at virtual objects in the virtual environment they find interesting, compared to objects that they have no interest in (or distractor objects and/or events). The disclosed subject matter provides systems and methods to use an hBCI to provide implicit reinforcement to an AI agent in a manner that incorporates user preferences into the control system ¶ [0028]-[0029], [0035]-[0036], [0043]-[0047], [0056], Figs. 3-5).
Alcaide discloses, engagement data in relation to specific operations and interactions occurring in said real-life context (In some embodiments, the video based eye-tracker 102 can be used to determine where a user is looking in their visual field by rapidly following the eye movements of the user in a two or three dimensional space. For example, provided the user has voluntary control of their eye-movements, the video based eye tracer 102 can be used to determine which subspaces in their visual field their eyes are “pointing to.” In other words, the video based eye-tracker 102 can use the user's eye-movement trajectories as a pointing control feature, revealing significant information about the subject's intent and behavior. In some embodiments, aspects of where in the visual space their attention focused, what stimulus they are focused upon, or what stimulus they responded to, can be used effectively in the BCI system 100 ¶ [0032]-[0033]).
Regarding claim 4, the combination of Sajda, Myrden and Alcaide discloses, wherein said real-life context comprises a cognitive probing event (Sajda: The present disclosure relates to systems and methods for providing a hybrid brain-computer-interface (hBCI) that can detect an individual's reinforcement signals (e.g., level of interest, arousal, emotional reactivity, cognitive fatigue, cognitive state, or the like) in response to objects, events and/or actions by an AI agent in an environment by generating implicit reinforcement signals for improving an AI agent controlling actions in the relevant environment, such as an autonomous vehicle ¶ [0005]. Also see ¶ [0020]-[0021]).
Regarding claim 5, the combination of Sajda, Myrden and Alcaide discloses, wherein adapting operation of said machine-learning algorithm is performed based on neural brain activity data acquired from more than one human participant participating with said real-life context (Sajda: collecting physiological response from multiple experimental subjects ¶ [0051]-[0055], Figs. 7A-7D).
Regarding claim 6, the combination of Sajda, Myrden and Alcaide discloses, wherein an aspect of said real-life context with which said human participant is momentarily implicitly involved with is identified from human bio-signal data of said human participant acquired by at least one bio-signal sensor (Sajda The disclosed systems and methods provide a hybrid brain-computer-interface (hBCI) that can detect an individual's implicit reinforcement signals (e.g., level of interest, arousal, emotional reactivity, cognitive fatigue, cognitive state, or the like) in response objects, events, and/or actions in an environment by generating implicit reinforcement signals for improving an AI agent controlling, by way of example only, a simulated autonomous vehicle ¶ [0020]-[0022], [0029], [0039]-[0040], [0051]-[0052], Fig. 3 and 6).
Regarding claims 7 and 17, the combination of Sajda, Myrden and Alcaide discloses, wherein said multi-modal operational data, said brain activity signals, and said neural brain activity data are processed in real-time or quasi real-time for dynamically adapting operation of said machine-learning algorithm (Sajda: real-time EEG-based inference and neural/physiological feedback used to influence AI behavior and reinforcement learning architecture also processes environmental state and action rewards ¶ [0020]-[0022], [0043]-[0047], [0065]-[0066], Figs. 3, 4, 9).
Regarding claim 8, the combination of Sajda, Myrden and Alcaide discloses, wherein said multi-modal operational data comprises comprise at least one of physical data and virtual data originating from said real-life context, including data relating to technological states and technological state changes of at least one device operating in said real-life context, comprising at least one of device input states, device output states, device operational states, device game states, computer--aided design states, computer--simulated design states, computer peripheral device states, computer--controlled machinery states, and corresponding state changes (Sajda: In some embodiments, a deep reinforcement AI agent can receive, as inputs, physiological signals from the human user and driving performance information associated with the simulated vehicle in the virtual environment. In some embodiments, the hBCI can also include a head-mounted display (e.g., a commercially available Oculus Rift, HTC Vive, etc.) and a plurality of actuators. The deep reinforcement AI learner can transmit instructions to the actuators in the hBCI's head-mounted display to perform certain actions ¶ [0025]-[0029], [0037]-[0038], [0043]-[0046] and Figs. 2, 5A-5B).
Regarding claims 9 and 18, the combination of Sajda, Myrden and Alcaide discloses, wherein said machine-learning algorithm operates a reinforcement machine-learning algorithm comprising a reward function, and wherein said assessed at least one type of implicit neurophysiological response provides at least part of said reward function (Sajda: In some embodiments, the reinforcement learning module can train the AI agent to navigate the environment. For example, a deep reinforcement learning paradigm can be used that optimizes the function for learning the correct action under a given state S using the equation: By implementing this form of learning, the reward value can be adjusted to combine explicit reinforcement of driving performance with physiologically derived reinforcement to influence the behavior of the AI-controlled simulated autonomous vehicle ¶ [0044]-[0045], [0050]-[0053], Figs. 3, 5B, and 6).
Regarding claims 10 and 20, the combination of Sajda, Myrden and Alcaide discloses, wherein said machine-learning algorithm comprises an artificial intelligence data-processing algorithm (Sajda: Furthermore, the systems and methods disclosed can use neural, behavioral, and/or physiological signals to inform deep reinforcement learning based AI systems to enhance user comfort and/or trust in automation ¶ [0005]-[0007]. Also see ¶ [0020]-[0022], [0033]-[0036], [0043]-[0046], [0062]-[0065], Figs. 2-3).
Regarding claim 11, the combination of Sajda, Myrden and Alcaide discloses, identifying, by said information processing device, an aspect of said real-life context based on said acquired engagement data (Sajda: identifying objects of interest using physiological classification ¶ [0035]-[0036], [0040], [0051]-[0052], Figs. 3, and 5);
acquiring, by said information processing device, [multi-modal] operational data pertaining to said identified aspect (Sajda: processing environmental object features and image derived information ¶ [0035], [0040], [0045]-[0046]);
multimodal (Alcaide: In some embodiments, the video based eye-tracker 102 relies on tracking the user's pupil and a first-surface corneal reflection (CR) of an illumination source with the use of a head-mounted eye tracking video camera to image the user's eye. The positional difference between these two features can be used to determine the observer's eye-in-head orientation. Some example head mounted eye-tracking devices that can be used as the video based eye-tracker 102, are shown in FIGS. 2A and 2B and are available from SenseMotoric Instruments, Tobii Eye Tracking, and Pupil-labs among other commercial vendors. For example, the illustration in FIG. 2B shows a front and rear view of a video based eye-tracker, with the left and right eye cameras indicated. The cameras can be connected to each other, through a wired or wireless connection. The video based eye-tracker can also include an additional scene camera that captures the uses' field of view ¶ [0033]);
acquiring, by said passive brain-computer interface, electrophysiological brain activity signals of said human participant participating with said real-life context pertaining to said identified aspect (Sajda: EEG responses associated with particular objects and events ¶ [0029], [0040], [0051]-[0052]);
providing, by said passive brain-computer interface, from said acquired electrophysiological brain activity signals, neural brain activity data to said identified aspect (Sajda: FIG. 2 is a block diagram illustrating a system level diagram of the disclosed hBCI system 200. The hBCI system 200 can include one or more sensory input devices 202 as shown in FIG. 1, a machine learning module 210 that can process input signals from sensory input device(s) 202 to reinforce driving behavior using a deep reinforcement network, and an environment module 204 that can generate a virtual environment in which the AI agent can drive a simulated autonomous vehicle using the determined reinforced driving behavior ¶ [0033], also see ¶ [0035]-[0036], [0040]-[0042], and Fig. 3), and
assessing, by said information processing device, from said neural brain activity data pertaining to said identified aspect, at least one type of implicit neurophysiological response of said human participant pertaining to said identified aspect (correlating physiological reactions with environmental events to infer the user’s cognitive state and interest ¶ [0029], [0040], [0047], and Fig. 4), and
adapting, by said information processing device, at least one of an analysis and inferencing operation of said machine-learning algorithm and aspects of said real-life context based on said assessed at least one type of implicit neurophysiological response of said human participant (Sajda: inferred physiological responses to influence AI learning and vehicle-control behavior ¶ [0043]-[0047], [0056], Fig. 3-4).
Regarding claim 12, the combination of Sajda, Myrden and Alcaide discloses, wherein said machine-learning algorithm is comprised of at least one algorithm configured for at least one of sensing, performing, acquiring, processing, and pre-processing of at least one of neural brain activity data, [multi-modal] operational data, context aspect identification, implicit neurophysiological response data, human bio-signal data and at least one other algorithm configured for operating with said real-life context and at least one of a control operation and inference operation of which performed in relation to said real-life context is adapted based on said assessed at least one type of implicit neurophysiological response of said human participant and said linked [multi-modal] operational data (Sajda: cooperating hBCI, TAG, and reinforcement modules; TAG computer vision algorithm processes object features and classifications in the environment; adapting an AI agent using physiological reinforcement and environmental information ¶ [0033]-[0036], [0043]-[0047], [0056], [0066]; The disclosed systems and methods provide a hybrid brain-computer-interface (hBCI) that can detect an individual's implicit reinforcement signals (e.g., level of interest, arousal, emotional reactivity, cognitive fatigue, cognitive state, or the like) in response objects, events, and/or actions in an environment by generating implicit reinforcement signals for improving an AI agent controlling, by way of example only, a simulated autonomous vehicle ¶ [0020]. The disclosed system can physiologically correlate such labeling to construct a hybrid brain-computer interface (hBCI) system for efficient navigation of the three-dimensional virtual environment. For example, neural and ocular signals reflecting subjective assessment of objects in a three-dimensional virtual environment can be used to inform a graph-based learning model of that virtual environment, resulting in an hBCI system that can customize navigation and information delivery specific to the passenger's interests. The physiological signals that were naturally evoked by virtual objects in this task can be classified by an hBCI system which can include a hierarchy of linear classifiers, as illustrated in FIG. 3. The hBCI system's linear classifiers along with Computer Vision (CV) features of the virtual objects can be used as inputs to a CV system (e.g., TAG module) ¶ [0040]-[0042], and Figs. 2-3).
multimodal (Alcaide: In some embodiments, the video based eye-tracker 102 relies on tracking the user's pupil and a first-surface corneal reflection (CR) of an illumination source with the use of a head-mounted eye tracking video camera to image the user's eye. The positional difference between these two features can be used to determine the observer's eye-in-head orientation. Some example head mounted eye-tracking devices that can be used as the video based eye-tracker 102, are shown in FIGS. 2A and 2B and are available from SenseMotoric Instruments, Tobii Eye Tracking, and Pupil-labs among other commercial vendors. For example, the illustration in FIG. 2B shows a front and rear view of a video based eye-tracker, with the left and right eye cameras indicated. The cameras can be connected to each other, through a wired or wireless connection. The video based eye-tracker can also include an additional scene camera that captures the uses' field of view ¶ [0033]).
Regarding claim 19, the combination of Sajda, Myrden and Alcaide discloses, wherein the said reinforcement machine-learning algorithm comprises a deep reinforcement machine-learning algorithm comprising a reward function, and wherein said assessed at least one type of implicit neurophysiological response provides at least part of said reward function (Sajda: the systems and methods disclosed can use neural, physiological, or behavioral signatures to inform deep reinforcement learning based AI systems to enhance user comfort and trust in automation [Abstract], ¶ [0005], [0008], [0020]-[0030]).
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MOHAMMAD S ROSTAMI whose telephone number is (571)270-1980. The examiner can normally be reached Mon-Fri From 9 a.m. to 5 p.m..
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Boris Gorney can be reached at (571)270-5626. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
10/10/2026
/MOHAMMAD S ROSTAMI/ Primary Examiner, Art Unit 2154
