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 .
This is a Non-Final Office Action in response to application 18/620,099 entitled "ADVERSARIAL IMITATION LEARNING ENGINE FOR ACTION RISK ESTIMATION BASED ON SENSOR DATA" filed on February, 20, 2026, with claims 1 to 20 pending.
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.
Status of Claims
Claims 1, 6, 10, 15, and 20 have been amended and are hereby entered.
Claims 1-20 are pending and have been examined.
Response to Amendment
The amendment filed February 20, 2026, has been entered. Claims 1-20 remain pending in the application. Applicant’s amendments to the Specification, Drawings, and/or Claims have been noted in response to the Final Office Action mailed November 28, 2025.
Examiner’s Note
Intended use language is generally given limited patentable weight. See MPEP 2114(II) ("A claim containing a 'recitation with respect to the manner in which a claimed apparatus is intended to be employed does not differentiate the claimed apparatus from a prior art apparatus’ if the prior art apparatus teaches all the structural limitations of the claim. Ex parte Masham, 2 USPQ2d 1647 (Bd. Pat. App. & Inter. 1987).”); see also MPEP 2103(C). Examples of claim limitations that are often found to precede intended use include “adapted to,” “capable of,” “sufficient to,” “whereby,” and “for.” The following limitations are interpreted as intended use limitations:
The examiner would like to note that the claims are replete with intended use, however, to provide compact prosecution, the examiner has provided the mapping and rejections.
Claims 1, 10, and 20: “…where the generator obfuscates the discriminator to enable the generator to generate the action sequences that mimic expert demonstrations…”
The Examiner determines the aforementioned intended use statements do not result in any structural nor manipulative difference between the claimed invention and the prior art. Therefore, the intended use statements are afforded limited patentable weight.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 1-20 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Independent claims 1, 10 and 20 were amended to require (claim 1): “controlling an autonomous vehicle, including changing steering direction of the autonomous vehicle, to avoid accidents through driving components of the autonomous vehicle.”
The original Specification does not support the full scope of this limitation. In the Remarks filed October 14, 2025, Applicant refers to Spec [0094], [0121-0138] and the Remarks of February, 20, 2026, cite [0069],[0119], [0156], [0177]; FIG. 1, 4-6, 10 and 12.
Paragraph [0094] merely states, “The output is a multi-faceted recommendation that not only proposes a sequence of actions aimed at achieving optimal operational performance but also quantifies the expected outcomes in the form of an estimated KPI and can execute corrective actions automatically (e.g., turn on/off components of an industrial system/vehicle, provide AI navigation and driving assistance to improve driving safety, etc.). This multi output mechanism ensures that the operational decisions are both proactive and grounded in robust analytical predictions, thus exemplifying a tangible advancement in the realm of industrial automation and risk management within the insurance industry. This architecture underpins the system's ability to adaptively optimize actions in pursuit of KPI maximization (e.g., carbon offset minimization), as well as enhancing risk assessment actions, all while addressing the challenges of large action sequence spaces, numerical data processing, and uncertain reward structures, in accordance with aspects of the present invention.” Paragraphs [0121-0138] describes “real-world driver’s actions” and “driver’s action sequence generation task.” Paragraphs [0121-0138] do not describe controlling an autonomous vehicle through driving components by processing multiple trajectories. These paragraphs describe assessing a driver’s score, a driver’s actions, and a driver’s behavior.
The Specification [0094] is the closest support but merely states automatically turning on/off a component of a vehicle, but there is no support for autonomous vehicle control and there is no support for controlling an autonomous vehicle through driving components. Further, [0094] states this automatic control is for “achieving optimal operational performance” and does not support the claim language “to avoid accidents.”
Similarly, Specification [0156] “a testing dataset can be used as input, and can be a comprehensive array of sensor data collected from vehicles during operation. This dataset is rich in detail, providing a temporal sequence of events and actions that accurately represent real-world driving conditions and behaviors. The data can include, but is not limited to, speed variations, braking intensity, steering angles, and turn signal usage, which are useful for the subsequent risk assessment process. In block 904, the system can leverage trained models that have been rigorously developed and validated using substantial historical datasets. These models can incorporate advanced algorithms designed to detect, analyze, and interpret complex driving patterns, making them robust tools for evaluating real-time vehicle sensor data. The models can be fine- tuned to identify nuances in driving behaviors that contribute to risk profiles, in accordance with aspects of the present invention.” It states that testing datasets include, “speed variations, braking intensity, steering angles, and turn signal usage” but never states the vehicle’s steering is controlled as a result.
Therefore, the claims are rejected.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
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.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 5, 8, 9, 10, 14, 17, 18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Akino (“SYSTEM AND METHOD FOR AUTOMATED TRANSFER LEARNING WITH DOMAIN DISENTANGLEMENT”, U.S. Publication Number: 20230162023 A1), in view of Yanosy (“ARTIFICIAL INTELLIGENCE ENHANCED KNOWLEDGE FRAMEWORK”, U.S. Publication Number: 20240296352 A1),in view of Gaurav (“SYSTEMS AND METHODS TO LEARN CONSTRAINTS FROM EXPERT DEMONSTRATIONS”, U.S. Publication Number: 20230376749 A1),in view of Edwards (“ADVERSARIAL LEARNING OF PRIVACY PROTECTION LAYERS FOR IMAGE RECOGNITION SERVICES”, U.S. Publication Number: 20190188830 A1).
Regarding Claim 1,
Akino teaches,
A computer-implemented method,
(Akino [0035] the AI model performs a classification task)
comprising: monitoring sensors to collect sensor data related to a state of a plurality of components;
(Akino [0035] The observation data are tensor formats with at least one axis to represent numerous signals and sensor data
Akino [0039] representing a spatio-temporal spectrogram from multiple-channel sensors over a measurement time.
Akino [0056] Given sensor measurements such as media data, physical data and physiological data)
processing, by a computing system, the sensor data to generate an action sequence …for each of the components;
(Akino [0035] schematic of an artificial intelligence (AI) model, which provides an inference to identify a task label Y from an observation data …The observation data are tensor formats with at least one axis to represent numerous signals and sensor data
Akino [0039] data signals with a pair of X and Y are bundled as a whole batch of dataset for training the AI model, and they are called training data or training dataset for supervised learning
Akino [0117] acts performed as part of the method may be ordered in any suitable way. Accordingly, embodiments may be constructed in which acts are performed)
generating, by the computing system, a … score
(Akino [0074] The MIGE method uses score function ... where several kernel-based score estimators are known, e.g.: ...Sliced Score Matching (SSM). The kernel-based score estimators ... chosen depending on the datasets.
Akino [0107] provide smaller loss values ... cross entropy...negative log-likelihood, absolute error, ...clustering loss, divergence, hinge loss, H...Multiple loss functions are further weighted)
for the action sequence using a Generative Adversarial Network (GAN),
(Akino [0004] The concept of adversarial learning was considered in Generative Adversarial Networks (GAN), and has been adopted into myriad applications)
wherein the GAN includes a generator that generates action sequences and a discriminator that distinguishes low-risk action sequences from high-risk action sequences in accordance with a threshold
(Akino [0039] data signals with a pair of X and Y are bundled as a whole batch of dataset for training the AI model, and they are called training data or training dataset for supervised learning
Akino [0117] acts performed as part of the method may be ordered in any suitable way. Accordingly, embodiments may be constructed in which acts are performed
Akino [0091] alternating optimization the discriminator and encoder are optimized in separate steps, using the two loss terms
Akino [0097] the model assigns a probability to any class which is above a threshold
Akino [0087] we train the adversary module's parameters to minimize a standard cross entropy loss for its prediction task.
Akino [0063] we can identify ... the nuisance variations S for a given generative model. ...Under a constrained risk minimization framework, there are multiple types of such censoring modes
Akino [0094] When the predicted score is above a threshold...The model is trained to make its prediction on the strongly-augmented version match the pseudo-label via a cross-entropy loss minimization.
Examiner notes the AL models distinguish high and low risk (minimize a standard cross entropy, nuisance variations, censoring, all under a constrained risk minimization framework) )
Akino does not teach risk score; simulates future driving behavior based on an action space of action vectors that adhere to a multivariate distribution by assigning distinct values to each action types using a transformer-based policy network that utilizes an Adversarial Imitation Learning Engine (AILE); the generator and discriminator engage in a min-max game based on a cross-entropy loss where the generator obfuscates the discriminator to enable the generator to generate the action sequences that mimic expert demonstrations; associating, by the computing system, the low-risk action sequences with components in the plurality of components based on the risk score; and communicating, by the computing system, a status of the low-risk action sequences; and co controlling an autonomous vehicle, including changing steering direction of the autonomous vehicle, to avoid accidents through driving components of the autonomous vehicle by simultaneously processing multiple trajectories generated with the transformer-based policy network based on the status of the low-risk action sequences.
Yanosy teaches,
risk score
(Yanosy [0012] total risk level, opportunity level, etc.
Yanosy [0187] the information provided regarding the opportunity and risk...indicating that the risk or opportunity is low, medium, or high, a numerical score can be provided, a percentile score can be provided similar to the overall percentile rating 612, or additional quantitative categories can be provided (e.g. very low, low, medium, high, very high).)
that adhere to a multivariate distribution by assigning distinct values to each action types using a transformer-based policy network;
(Yanosy [0099] the transformer type model can employ an encoder stack and a decoder stack
Yanosy [0213] a linear regression analysis on a larger scale where three or more different variables are being analyzed…. certain variable has a high correlation with the actual characteristic)
associating, by the computing system, the low-risk action sequences with components in the plurality of components based on the risk score;
(Yanosy [0192] the percentile rating is in the 50.sup.th percentile, and the various risk metrics and opportunity metrics... the percentile rating is in the 80.sup.th percentile, and the various risk metrics
Yanosy [0174] improvement module 416 can provide suggestions for reducing risk...can identify various responses which have a factors with a low score and can indicate to the respondent certain actions that can be taken...can beneficially increase the relevant scoring for a certain assessment to potentially improve the final decision to an improved category)
and communicating, by the computing system, a status of the low-risk action sequences.
(Yanosy [0139] dependent intrinsic characteristic 308 to acquire sufficient information through their answers that enable additional insights about the status/impact of the assessment object
Yanosy [0145] a default value for a decision 322 can be “NotSatisfied” so that the assessment does not support a positive decision. The decision logic 320 can assert its interpretation... class to a “Satisfied” or “NotSatisfied” value
Yanosy [0172] can determine final innovation decision results
Yanosy [0064] screens that can be presented on a display)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the knowledge framework teachings of Yanosy “for generating a knowledge framework that integrates knowledge obtained from a machine learning unit.” (Yanosy [0002]). The modification would have been obvious, because it is merely applying a known technique (i.e. knowledge framework) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “a universal assessment theory model that can be refined for a specific assessment object being assessed, which can be of various types” Yanosy [0002])
Yanosy does not teach simulates future driving behavior based on an action space of action vectors; utilizes an Adversarial Imitation Learning Engine (AILE); the generator and discriminator engage in a min-max game based on a cross-entropy loss where the generator obfuscates the discriminator to enable the generator to generate the action sequences that mimic expert demonstrations; controlling an autonomous vehicle, including changing steering direction of the autonomous vehicle, to avoid accidents through driving components of the autonomous vehicle by simultaneously processing multiple trajectories generated with the … based on the status of the low-risk action sequences.
Gaurav teaches,
simulates future driving behavior based on an action space of action vectors;
(Gaurav [0136] represents actions taken by a reinforcement learning agent in the simulated environment
Gaurav [0037] a data sample provided to a trained machine learning model which will infer (i.e. predict) an output
Gaurav [0141] configured to manage different aspects of the driving process, such as prediction, perception, planning, etc. The planning component is usually divided into three parts: mission planning ... behavior planning ...and motion planning
Gaurav [0026] An MDP is defined as a tuple (S, A, p, μ, r, γ), wherein S is the state space, A is the action space, p(·|s, α) are the transition probabilities over the next states given the current states and current action)
utilizes an Adversarial Imitation Learning Engine (AILE);
(Gaurav [0165] The first baseline approach was GAIL-Constraint, i.e. Generative Adversarial Imitation Learning: an imitation learning method that can be used to learn a policy that mimics the expert policy)
controlling an autonomous vehicle, including changing steering direction of the autonomous vehicle, to avoid accidents through driving components of the autonomous vehicle by simultaneously processing multiple trajectories generated with the … based on the status of the low-risk action sequences.
(Gaurav [0144] a planning module 530 that generates a control signal
Gaurav [0141] motion planning (i.e., generating low-level control signals, such as immediate steering and immediate acceleration, to execute a high-level driving action).
Gaurav [Claim 18] computing the second utility by applying the current constraint function to the plurality of agent trajectories
Gaurav [0003] An example of these function in autonomous driving ...defines movements that are not allowed such as driving off the road, getting into a collision, accelerating toward a red traffic light, etc.
Gaurav [0013] allow some risk-taking behavior in order to achieve the objective at hand...which are on average satisfied across a set of all trajectories
Gaurav [0025] The threshold may be a lower limit...of absolute magnitude, or any other limit.)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the vehicle control teachings of Gaurav for “a planning module 530 that generates a control signal.” (Gaurav [0144]). The modification would have been obvious, because it is merely applying a known technique (i.e. vehicle control) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “Reinforcement learning is a machine learning technique oriented toward solving planning problems.” Gaurav [0003])
Gaurav does not teach the generator and discriminator engage in a min-max game based on a cross-entropy loss where the generator obfuscates the discriminator to enable the generator to generate the action sequences that mimic expert demonstrations;
Edwards teaches,
the generator and discriminator engage in a min-max game
(Edwards [0006] proposed technique of Generative Adversarial Networks (GANs) repurposes the min/max paradigm from game theory to generate images in an unsupervised manner. The GAN framework comprises a generator and a discriminator)
based on a cross-entropy loss
(Edwards [0060] The image recognition neural network may be trained using one loss function, e.g., Loss 1 in this example, which may be a cross-entropy loss.)
where the generator obfuscates the discriminator to enable the generator
(Edwards [0023] With the adversarial neural network based learning framework, a generator is provided that is a pre-processor that performs some degree of image blurring or obfuscation and provides the modified image to one or more discriminators.
Edwards [0039] the generator may select between different types of obfuscation mechanisms based on the discriminator output and output from the image recognition service to achieve a lowest possible loss (highest accuracy) of the adversarial network with the generator (pre-processor) being trained to achieve a highest loss (lowest accuracy) while achieving high accuracy output of the image recognition service.)
to generate the action sequences
(Edwards [0027] for accomplishing and/or performing the actions, steps, processes, etc.,)
that mimic expert demonstrations
(Edwards [0043] A cognitive system performs one or more computer-implemented cognitive operations that approximate a human thought process as well as enable people and machines to interact in a more natural manner so as to extend and magnify human expertise and cognition.)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the min-max game teachings of Edwards for a “min/max paradigm from game theory to generate images in an unsupervised manner.” (Edwards [0006]). The modification would have been obvious, because it is merely applying a known technique (i.e. min-max game) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “where the generator acts as an adversary and tries to fool the discriminator by producing synthetic images based on a noise input, and the discriminator tries to differentiate synthetic images from true images.” Edwards [0006])
Regarding Claim 5,
Akino, Yanosy, Gaurav, and Edwards teach the sequence classification of Claim 1 as described earlier.
Akino teaches,
wherein monitoring the sensors
(Akino [0035] The observation data are tensor formats with at least one axis to represent numerous signals and sensor data)
Akino does not teach includes monitoring vehicles and the action sequences include driver actions.
Gaurav teaches,
includes monitoring vehicles and the action sequences include driver actions.
(Gaurav [Claim 1] obtaining: demonstration data representative of the demonstration, the demonstration data comprising a sequence of actions, each action being taken in the context of a respective state of a demonstration environment
Gaurav [0033] a “trajectory” may refer to a literal physical trajectory of the vehicle being driven by an agent)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the vehicle monitoring of Gaurav “comprises operating an autonomous driving system by operating a motion planner of the autonomous driving system in accordance with the constraint function.” (Gaurav [0051]). The modification would have been obvious, because it is merely applying a known technique (i.e. vehicle monitoring) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “Reinforcement learning is a machine learning technique oriented toward solving planning problems.” Gaurav [0003])
Regarding Claim 8,
Akino, Yanosy, Gaurav, and Edwards teach the sequence classification of Claim 1 as described earlier.
Akino teaches,
wherein processing the sensor data includes generating action sequences with discerning temporal correlations
(Akino [0035] an artificial intelligence (AI) model, which provides an inference to identify a task label Y from an observation data …The observation data are tensor formats with at least one axis to represent numerous signals and sensor data
Akino [0043] include a set of subject identifications, ...sensor states,...time and sensitivities
Akino [Claim 15] are mutually connected with a plural of … activation functions to pass a signal from the previous layers to the next layers sequentially.)
Akino does not teach using an adapted Transformer architecture to capture distinctions and dependencies within the action sequences.
Yanosy teaches,
using an adapted Transformer architecture to capturing distinctions and dependencies within the action sequences.
(Yanosy [0092] The transformer type model or architecture can be configured to process sequences of data
Yanosy [0093] making predictions and helps capture dependencies and relationships... allow the model to capture both local and global dependencies)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the knowledge framework teachings of Yanosy “for generating a knowledge framework that integrates knowledge obtained from a machine learning unit.” (Yanosy [0002]). The modification would have been obvious, because it is merely applying a known technique (i.e. knowledge framework) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “a universal assessment theory model that can be refined for a specific assessment object being assessed, which can be of various types” Yanosy [0002])
Regarding Claim 9,
Akino, Yanosy, Gaurav, and Edwards teach the sequence classification of Claim 1 as described earlier.
Akino teaches,
further comprising assigning … scores to individual actions of components
(Akino [0063] Under a constrained risk minimization framework
Akino [0074] The MIGE method uses score function ... where several kernel-based score estimators are known, e.g.: ...Sliced Score Matching (SSM). The kernel-based score estimators ... chosen depending on the datasets.
Akino [0107] provide smaller loss values ... cross entropy...negative log-likelihood, absolute error, ...clustering loss, divergence, hinge loss, H...Multiple loss functions are further weighted)
using a performance prediction neural network.
(Akino [0016] DNN blocks via a stochastic gradient optimization such that a task prediction is accurate)
Akino does not teach risk score;
Yanosy teaches,
risk score
(Yanosy [0012] total risk level, opportunity level, etc.
Yanosy [0187] the information provided regarding the opportunity and risk...indicating that the risk or opportunity is low, medium, or high, a numerical score can be provided, a percentile score can be provided similar to the overall percentile rating 612, or additional quantitative categories can be provided (e.g. very low, low, medium, high, very high).)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the knowledge framework teachings of Yanosy “for generating a knowledge framework that integrates knowledge obtained from a machine learning unit.” (Yanosy [0002]). The modification would have been obvious, because it is merely applying a known technique (i.e. knowledge framework) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “a universal assessment theory model that can be refined for a specific assessment object being assessed, which can be of various types” Yanosy [0002])
Claim 10 is rejected on the same basis as Claim 1.
Claim 14 is rejected on the same basis as Claim 5.
Claim 17 is rejected on the same basis as Claim 8.
Claim 18 is rejected on the same basis as Claim 9.
Claim 20 is rejected on the same basis as Claim 1.
Claim 2-4, 11-13, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Akino, Yanosy, Gaurav, and Edwards in view of Dalli (“EXPLAINABLE TRANSDUCER TRANSFORMERS”, U.S. Publication Number: 20230153599 A1).
Regarding Claim 2,
Akino, Yanosy, Gaurav, and Edwards teach the sequence classification of Claim 1 as described earlier.
Akino does not teach further comprising training the transformer-based policy network and the GAN using multi-head self-attention mechanisms to process sequential sensor inputs.
Yanosy teaches,
further comprising training the transformer-based policy network and the GAN ….using multi-head self-attention mechanisms to process sequential sensor inputs.
(Yanosy [0099] the transformer type model can employ an encoder stack and a decoder stack
Yanosy [0091] generative adversarial networks (GANs), transformers, and the like.
Yanosy [0092] self-attention and multi-head attention mechanisms
Yanosy [0096] effectively process sequential data
Yanosy [0166] configured to receive input data from one or more sensors)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the knowledge framework teachings of Yanosy “for generating a knowledge framework that integrates knowledge obtained from a machine learning unit.” (Yanosy [0002]). The modification would have been obvious, because it is merely applying a known technique (i.e. knowledge framework) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “a universal assessment theory model that can be refined for a specific assessment object being assessed, which can be of various types” Yanosy [0002])
Yanosy does not teach with the AILE; by integrating outputs of individual attention head layers to encapsulate temporal interdependencies across various timestamps.
Gaurav teaches,
with the AILE
(Gaurav [0165] The first baseline approach was GAIL-Constraint, i.e. Generative Adversarial Imitation Learning: an imitation learning method that can be used to learn a policy that mimics the expert policy)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the GAIL-Constraint teachings of Gaurav for “Generative Adversarial Imitation Learning” (Gaurav [0165]). The modification would have been obvious, because it is merely applying a known technique (i.e. GAIL-Constraint) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “an imitation learning method that can be used to learn a policy that mimics the expert policy.” Gaurav [0165])
Gaurav does not teach by integrating outputs of individual attention head layers to encapsulate temporal interdependencies across various timestamps.
Dalli teaches,
by integrating outputs of individual attention head layers to encapsulate temporal interdependencies across various timestamps.
(Dalli [0007] architecture, may include two sub-layers. The first sub-layer may include a Multi-Head Attention component
Dalli [0008] The first sub-layer consists of a Masked Multi-Head Attention component
Dalli [0022] generate a contextual representation of the temporal dependencies....for each time point.)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the attention head layers teachings of Dalli where “The first sub-layer consists of a Masked Multi-Head Attention component” (Dalli [0008]). The modification would have been obvious, because it is merely applying a known technique (i.e. attention head layers) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “an attention-based architecture that offers state-of-the-art results in various fields.” Dalli [0005])
Regarding Claim 3,
Akino, Yanosy, Gaurav, Edwards, and Dalli teach the sequence classification of Claim 2 as described earlier.
Akino does not teach wherein training includes pre-training on a labeled dataset, the sensor data from known low-risk action sequences to simulate action sequences.
Yanosy teaches,
wherein training includes pre-training on a labeled dataset, the sensor data from known low-risk action sequences to simulate action sequences.
(Yanosy [0098] The models can be pre-trained and trained on massive data corpora
Yanosy [0091] including supervised learning techniques...where the model is trained on a partially labeled dataset)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the knowledge framework teachings of Yanosy “for generating a knowledge framework that integrates knowledge obtained from a machine learning unit.” (Yanosy [0002]). The modification would have been obvious, because it is merely applying a known technique (i.e. knowledge framework) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “a universal assessment theory model that can be refined for a specific assessment object being assessed, which can be of various types” Yanosy [0002])
Regarding Claim 4,
Akino, Yanosy, Gaurav, Edwards, and Dalli teach the sequence classification of Claim 3 as described earlier.
Akino teaches,
for real-time generation of … scores and distinguishing between real and synthetic action sequences.
(Akino [0063] Under a constrained risk minimization framework
Akino [0074] The MIGE method uses score function ... where several kernel-based score estimators are known,
Akino [0006] scoring methods
Akino [0083] comparing its own average autoencoder loss on real data and fake data... estimate of the Wasserstein-1 distance between the real and fake autoencoder losses
Akino [0091] while a discriminator network tries to distinguish real and fake data samples.)
Akino does not teach risk score; wherein training includes deploying the trained transformer-based policy network and the GAN to process incoming unlabeled sensor data.
Yanosy teaches,
risk score
(Yanosy [0012] total risk level, opportunity level, etc.
Yanosy [0187] the information provided regarding the opportunity and risk...indicating that the risk or opportunity is low, medium, or high, a numerical score can be provided, a percentile score can be provided similar to the overall percentile rating 612, or additional quantitative categories can be provided (e.g. very low, low, medium, high, very high).)
wherein training includes deploying the trained transformer-based policy network and the GAN to process incoming unlabeled sensor data
(Yanosy [0099] the transformer type model can employ an encoder stack and a decoder stack
Yanosy [0091] generative adversarial networks (GANs), transformers, and the like.
Yanosy [0091] including supervised learning techniques...where the model is trained on a partially labeled dataset
Yanosy [0091] analyze and identify patterns in unlabeled data
Yanosy [0166] configured to receive input data from one or more sensors)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the knowledge framework teachings of Yanosy “for generating a knowledge framework that integrates knowledge obtained from a machine learning unit.” (Yanosy [0002]). The modification would have been obvious, because it is merely applying a known technique (i.e. knowledge framework) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “a universal assessment theory model that can be refined for a specific assessment object being assessed, which can be of various types” Yanosy [0002])
Claim 11 is rejected on the same basis as Claim 2.
Claim 12 is rejected on the same basis as Claim 3.
Regarding Claim 13,
Akino, Yanosy, Gaurav, Edwards, and Dalli teach the sequence classification of Claim 12 as described earlier.
Akino teaches,
wherein the GAN processes
(Akino [0004] The concept of adversarial learning was considered in Generative Adversarial Networks (GAN), and has been adopted into myriad applications)
incoming unlabeled data
(Akino [0049] for unsupervised feature extraction
Akino [Claim 16] unsupervised settings)
and distinguishes between real and generated action sequences.
(Akino [0083] comparing its own average autoencoder loss on real data and fake data... estimate of the Wasserstein-1 distance between the real and fake autoencoder losses
Akino [0091] while a discriminator network tries to distinguish real and fake data samples.)
Claim 19 is rejected on the same basis as Claim 2.
Claims 6 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Akino, Yanosy, Gaurav, and Edwards in view of Jaganathan (“ARTIFICIAL INTELLIGENCE-BASED SEQUENCING”, U.S. Publication Number: 20200302224 A1).
Regarding Claim 6,
Akino, Yanosy, Gaurav, and Edwards teach the sequence classification of Claim 1 as described earlier.
Akino does not teach wherein the cross-entropy loss includes an objective function for the discriminator that utilizes a maximum of an expected value of a logarithm of an output of a multiple-layer perceptron (MLP) for a state and an action.
Jaganathan teaches,
wherein the cross-entropy loss
(Jaganathan [1938] wherein the loss function is custom-weighted binary cross-entropy loss and the error is minimized)
includes an objective function for the discriminator that utilizes a maximum of an expected value
(Jaganathan [0409] the objective function is set to reconstruct a segmentation mask using a loss function
Jaganathan [0496] The training 2800 includes iteratively optimizing a loss function
Jaganathan [1320] variance can be the discrepancy between an expected value and a measured value.
Jaganathan [0813] based on the likelihoods (e.g., the base with the maximum likelihood is selected
Jaganathan [1311] the values include near maximum.... values can include local maximum....values include only absolute maximum)
of a logarithm of an output of a multiple-layer perceptron (MLP) for a state and an action;
(Jaganathan [1006] quality scores are logarithmically based
Jaganathan [0665] the neural network-based base caller 1514 is a multilayer perceptron (MLP).)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the objective function teachings of Jaganathan “objective function is set to reconstruct a segmentation mask using a loss function.” (Jaganathan [0409]). The modification would have been obvious, because it is merely applying a known technique (i.e. objective function) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “training 2800 includes iteratively optimizing a loss function that minimizes error 2806 between the decay map 1716 and the ground truth decay map 1204, and updating parameters of the regression model 2600 based on the error” Jaganathan [0496])
Regarding Claim 15,
Akino, Yanosy, Gaurav, and Edwards teach the sequence classification of Claim 10 as described earlier.
Akino does not teach wherein the cross-entropy loss includes an objective function for the discriminator that utilizes a minimum of an expected value of a logarithm of an output of a multiple-layer perceptron (MLP) for a state and an action.
Jaganathan teaches,
wherein the cross-entropy loss
(Jaganathan [1938] wherein the loss function is custom-weighted binary cross-entropy loss and the error is minimized)
includes an objective function for the discriminator that utilizes a minimum of an expected value
(Jaganathan [0409] the objective function is set to reconstruct a segmentation mask using a loss function
Jaganathan [0496] The training 2800 includes iteratively optimizing a loss function
Jaganathan [1320] variance can be the discrepancy between an expected value and a measured value.
Jaganathan [1311] the values include … near minimum values.... can include … local minimum values.... include only absolute … minimum values.)
of a logarithm of an output of a multiple-layer perceptron (MLP) for a state and an action;
(Jaganathan [1006] quality scores are logarithmically based
Jaganathan [0665] the neural network-based base caller 1514 is a multilayer perceptron (MLP).)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the objective function teachings of Jaganathan “objective function is set to reconstruct a segmentation mask using a loss function.” (Jaganathan [0409]). The modification would have been obvious, because it is merely applying a known technique (i.e. objective function) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “training 2800 includes iteratively optimizing a loss function that minimizes error 2806 between the decay map 1716 and the ground truth decay map 1204, and updating parameters of the regression model 2600 based on the error” Jaganathan [0496])
Claims 7 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Akino, Yanosy, Gaurav, and Edwards in view of Yan (“PARTIAL CALIBRATION TARGET DETECTION FOR IMPROVED VEHICLE SENSOR CALIBRATION”, U.S. Publication Number: 20210051317 A1).
Regarding Claim 7,
Akino, Yanosy, Gaurav, and Edwards teach the sequence classification of Claim 1 as described earlier.
Akino teaches,
cleaning …the sensor data by computing statistical measures on the sensor data;
(Akino [Claim 13] wherein the datasets include a combination of sensor measurements
Akino [Claim 9] multi-dimensional tensor projection with a plural of trainable linear filters or bilinear filters to convert lower-dimensional signals
Akino [0058] relies on the Bayes-Ball algorithm to facilitate an automatic pruning)
Akino does not teach with the AILE; filtering… the sensor data to remove unrelated data by employing a Pearson correlation coefficient computation.
Yanosy teaches,
filtering… the sensor data to remove unrelated data
(Yanosy [0166] configured to receive input data from one or more sensors
Yanosy [0041] the session data can be filtered by identifying data that is trustworthy and data that is untrustworthy.)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the data filtering teachings of Yanosy to “filter the machine learning data.” (Yanosy [Claim 4]). The modification would have been obvious, because it is merely applying a known technique (i.e. data filtering) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “filter the machine learning data and the session data before use of the machine learning data and the session data in creating or enhancing the knowledge framework” Yanosy [Claim 4])
Yanosy does not teach with the AILE; by employing a Pearson correlation coefficient computation.
Gaurav teaches,
with the AILE.
(Gaurav [0165] The first baseline approach was GAIL-Constraint, i.e. Generative Adversarial Imitation Learning: an imitation learning method that can be used to learn a policy that mimics the expert policy,)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the GAIL-Constraint teachings of Gaurav for “Generative Adversarial Imitation Learning” (Gaurav [0165]). The modification would have been obvious, because it is merely applying a known technique (i.e. GAIL-Constraint) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “an imitation learning method that can be used to learn a policy that mimics the expert policy.” Gaurav [0165])
Gaurav does not teach by employing a Pearson correlation coefficient computation.
Yan teaches,
by employing a Pearson correlation coefficient computation.
(Yan [0081] the Pearson correlation coefficient)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the adversarial machine learning classification teachings of Akino to incorporate the sensor calibration teachings of Yan “for improving calibration of vehicle sensors.” (Yan [0001]). The modification would have been obvious, because it is merely applying a known technique (i.e. sensor calibration) to a known concept (i.e. adversarial machine learning classification) ready for improvement to yield predictable result (i.e. “improving calibration of vehicle sensors using partial calibration target detection” Yan [0001])
Claim 16 is rejected on the same basis as Claim 7.
Response to Remarks
Applicant's arguments filed on February 20, 2026, have been fully considered and Examiner’s remarks to Applicant’s amendments follow.
Response Remarks on Claim Rejections - 35 USC § 112
The Applicant states:
“Applicant has amended claims 1, 10, and 20 in a manner believed to have overcome the rejections under 35 U.S.C. § 112(a). Support for these amendments could be found in at least para. [0069],[0119], [0156], [0177]; FIG. 1, 4-6, 10 and 12 of the Specification as filed."
Examiner responds:
Applicant's Disclosure does not support the claim limitation of “controlling an autonomous vehicle, including changing steering direction of the autonomous vehicle, to avoid accidents through driving components of the autonomous vehicle”.
The original specification does not support the full scope of this limitation. In the Remarks filed February, 20, 2026, cite [0069],[0119], [0156], [0177]; FIG. 1, 4-6, 10 and 12.
The most relevant paragraph of Specification [0156] “a testing dataset can be used as input, and can be a comprehensive array of sensor data collected from vehicles during operation. This dataset is rich in detail, providing a temporal sequence of events and actions that accurately represent real-world driving conditions and behaviors. The data can include, but is not limited to, speed variations, braking intensity, steering angles, and turn signal usage, which are useful for the subsequent risk assessment process. In block 904, the system can leverage trained models that have been rigorously developed and validated using substantial historical datasets. These models can incorporate advanced algorithms designed to detect, analyze, and interpret complex driving patterns, making them robust tools for evaluating real-time vehicle sensor data. The models can be fine- tuned to identify nuances in driving behaviors that contribute to risk profiles, in accordance with aspects of the present invention.” It states that testing datasets include, “speed variations, braking intensity, steering angles, and turn signal usage” but never states the vehicle’s steering is controlled as a result.
The rejection under 35 USC § 112 remains.
Response Remarks on Claim Rejections - 35 USC § 101
The amended claim language states (claim 1), “controlling an autonomous vehicle, including changing steering direction of the autonomous vehicle, to avoid accidents through driving components of the autonomous vehicle”
This limitation, if found to be supported by the Specification, would integrate the abstract idea into a practical application. However, there exists a Rejection Under 35 USC § 112(a) New Matter for this limitation. Temporarily, the Rejection Under 35 USC § 101 is lifted.
Response Remarks on Claim Rejections - 35 USC § 103
Applicant's amendments required the application of new/additional prior art.
New prior art includes:
Edwards (“ADVERSARIAL LEARNING OF PRIVACY PROTECTION LAYERS FOR IMAGE RECOGNITION SERVICES”, U.S. Publication Number: 20190188830 A1).
Jaganathan (“ARTIFICIAL INTELLIGENCE-BASED SEQUENCING”, U.S. Publication Number: 20200302224 A1).
Applicant’s remarks regarding the rejection made under 35 USC § 103 are rendered moot by the introduction of additional prior art.
Therefore, the rejection under 35 USC § 103 remains.
Prior Art Cited But Not Applied
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Lyu (“AUTONOMOUS DRIVING METHODS AND SYSTEMS”, U.S. Publication Number: 20240160945 A1) proposes training a deep reinforcement learning model for autonomous control of a machine, such as autonomous vehicles, the model being configured to output, by a policy network, an agent action in response to input of state information and a value function, the agent action representing a control signal for the machine. The method comprises minimizing a loss function of the policy network; wherein the loss function of the policy network comprises an autonomous guidance component and a human guidance component (human intervention); and wherein the autonomous guidance component is zero when the state information is indicative of a human input signal.
Mueck (“ARTIFICIAL INTELLIGENCE REGULATORY MECHANISMS”, U.S. Publication Number: 20240273411 A1) proposes mechanisms for enforcing compliance with artificial intelligence (AI) and machine learning (ML) regulatory frameworks. The AI regulatory enforcement mechanisms are capable of testing AI systems for quality, accuracy, and robustness, as well as for compliance with AI regulatory requirements. AI regulatory enforcement mechanisms provide restrictions and safeguards by controlling actions of AI systems and/or other components to prevent erroneous or biased AI system predictions from being used, and potentially causing harm to individuals or objects. The AI regulatory enforcement mechanisms ensure that AI systems function in ways that are secure, trustworthy, and ethical.
Mustikovela (“TRAINING AND INFERENCING USING A NEURAL NETWORK TO PREDICT ORIENTATIONS OF OBJECTS IN IMAGES”, U.S. Publication Number: 20210150757 A1) provides identify orientations of objects within images. In at least one embodiment, one or more neural networks are trained to identify an orientations of one or more objects based, at least in part, on one or more characteristics of the object other than the object's orientation.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CHINEDU EKECHUKWU whose telephone number is (571)272-4493. The examiner can normally be reached on Mon-Fri 9 AM ET to 3:30 PM ET.
Examiner interviews are available via telephone 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, Christine Tran, can be reached on (571) 272-8103. 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.
/C.E./Examiner, Art Unit 3695
/CHRISTINE M Tran/Supervisory Patent Examiner, Art Unit 3695
Prosecution Insights