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
This action is in reply to the application filed 2024 August 20.
Claims 1-20 are currently pending and have been examined.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 20 August 2024 has been considered by the examiner and an initialed copy of the IDS is hereby attached.
The examiner reminds Applicant of their duty to disclose any information that that may be relevant to the patentability of their invention. Relevant information includes references and prior art by the inventors that may be relevant to the claimed invention.
Claim Objections
Claims 1, 9-17 are objected to because of the following informalities:
Claims 1 recites “using Ne elite path” in line 10. The examiner recommends amending to recite “using an Ne elite path”. Claims 9 and 17 have a similar recitation and are rejected for the same reason.
Claims 10-16 recites “the method of” and depend directly or indirectly from claim 9. Claim 9 is a system claim. It is not clear if the claims 10-16 are method claims or system claims. Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
The claims are replete with indefiniteness issues. While the examiner has attempted to identify all of them, any newly identified 112(a)/(b) rejections will not be considered a new basis of rejection because the Applicant initial submission has significant errors which cause the claims to be difficult to examine.
Claim 1 recites “the same path” in line 4. There is insufficient antecedent basis for this limitation in the claim. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “creating a samples array” in line 11. Claim 1 previously recited “creating a samples array” in line 9. It is not clear if the samples array recited in line 11 is the same or different than that recited in line 9. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “creating a samples array” in line 11. Claim 1 previously recited “creating a samples array” in line 9. It is not clear if there is two separate steps of creating the samples array or if there is a single step of creating the samples array. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “the sample array” in line 11. There is insufficient antecedent basis for this limitation in the claim. The examiner notes that “a samples array” was previously recited. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “evaluating a sample path” in line 18. Claim 1 previously recited “evaluating a sample path” in line 9. It is not clear if the sample path of line 18 is the same or different than that recited in line 9. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “evaluating a sample path” in line 18. Claim 1 previously recited “evaluating a sample path” in line 9. It is not clear if there are two separate steps of evaluating a sample path or if there is a single step of evaluating the sample path. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “using Ne elite path” in line 10. There is insufficient antecedent basis for this limitation in the claim. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “the sample array” in line 11. There is insufficient antecedent basis for this limitation in the claim. The examiner notes that “a samples array” was previously recited, but not “a sample array”. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “ wherein the custom score function asserts a potential position and a potential acceleration to identify a no collision” in lines 18-19. It is not clear how a function can “assert”. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “the perturbation-adversary” in lines 19-20. There is insufficient antecedent basis for this limitation in the claim. It is not clear if the perturbation adversary is the same or different than the adversary vehicle. Claim 17 has a similar recitation and is rejected for the same reasons.
Claim 1 recites “a small position and acceleration distance”. It is not clear what is meant by a small position and acceleration distance. First, it is not clear if the term small modifies both the position and the acceleration distance. Second it is not clear what is an acceleration distance. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Further, the term “small” in claim 1 is a relative term which renders the claim indefinite. The term “small” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “sorting the sample path” in line 23. Claim 1 previously recited “sorting a sample path” in lines 9-10. It is not clear if there are two separate steps of sorting a sample path or if there is a single step of sorting the sample path. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “the sample paths” in line 24. There is insufficient antecedent basis for this limitation in the claim. Claim 1 previously recited “a sample path” and “the sample path”, but did not recite “sample paths”. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “using Ne elite paths” in line 24. There is insufficient antecedent basis for Ne elite paths in the claim. Further, it is not clear if the Ne elite paths recited in line 24 is the same or different than the Ne elite path recited in line 10 of the claim. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “using Ne elite paths” in line 25. Claim 1 recites “using Ne elite path” in line 10. It is not clear if there are two separate steps of using Ne elite path or if there is a single step of using the Ne elite. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 1 recites “repeating until Ne elite paths stabilizes” . It is not clear what is repeated until the Ne elite path stabilizes. Are all the steps of claim 1 repeated or is a subset of the steps recited in the claim repeated until the Ne elite path stabilizes. Claims 9 and 17 have similar recitations and are rejected for the same reasons. Further, claim 17 recites “repeating until the Ne elite paths stabilizes for a predetermined number of iterations” and includes significantly more steps. As noted above, with respect to claim 1, it is not clear what steps are repeated.
Claim 1 recites “repeating” until Ne elite paths stabilizes”. The term “stabilizes” in claim 1 is a relative term which renders the claim indefinite. The term “stabilizes” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. It is not clear what would be variance would be considered stable or not stable. Claims 9 and 17 have similar recitations and are rejected for the same reasons.
Claim 3 recites “a slight deviation”. The term “slight” in claim 3 is a relative term which renders the claim indefinite. The term “slight” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. It is not clear what deviation would be considered slight or not slight. Claims 11 and 17 have similar recitations and are rejected for the same reasons.
Claim 3 recites “a non-colliding path”. However claim 3 depends from claim 1 which recites “a no collision path”. It is not clear if the non-colliding path of claim 3 is the same or different than the no collision path of claim 1. Claims 11 and 17 have similar recitations and are rejected for the same reasons.
Claim 3 recites “different colliding or non-colliding paths”. There is insufficient antecedent basis for this limitation in the claim. The examiner notes that claim 3 depends from claim 1 which recites “a collision path” in line 19. It is not clear if the colliding paths are the same or different than that of the collision path recited in claim 1. Claims 11 and 17 have similar recitations and are rejected for the same reasons.
Claim 4 recites “the vanilla adversary vehicle” . There is insufficient antecedent basis for this limitation in the claim.
Claim 4 recites “a potential position” and “a potential acceleration” in lines 5-6. Claim 4 depends from claim 1 which previously recites “a potential position” and a potential acceleration” in lines 18-19. It is not clear if these terms refer back to the limitations in claim 1, or if they are intending to introduce an additional potential position or potential acceleration. Claims 12 and 18 have similar recitations and are rejected for the same reasons.
Claim 4 recites “ wherein the custom score function asserts a potential position and a potential acceleration to identify a no collision”. It is not clear what is intended by this limitation. First it is not clear how a function can “assert”. Further, it is not clear if “a no collision” is “a no collision path”. It appears the current language of the claim is incomplete or otherwise grammatically incorrect. Claims 12 and 18 have similar recitations and are rejected for the same reasons.
Claim 4 recites “a small position and acceleration distance”. The examiner notes that claim 4 depends from claim 1 which previously recited “a small position and acceleration distance”. It is not clear if the “small position and acceleration distance” of claim 4 is the same or different than that of claim 1. Claims 12 and 18 have similar recitations and are rejected for the same reasons.
Claim 4 recites “a small position and acceleration distance between the adversary vehicle and independent adversary vehicle”. As noted above with respect to claim 1, it is not clear what is meant by a small position and acceleration distance. First, it is not clear if the term small modifies both the position and the acceleration distance. Second it is not clear what is an acceleration distance. Claims 12 and 18 have similar recitations and are rejected for the same reasons.
Further, the term “small” in claim 4 is a relative term which renders the claim indefinite. The term “small” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. Claims 12 and 18 have similar recitations and are rejected for the same reasons.
Claims 5-8 are method claims, however the claims do not positively recite method steps. For example claim 6 recites “ wherein realistic scenarios are simulated from one or more autonomous vehicle reporting databases”. It is unclear what his required by the scope of the claim. The recitation of “simulated” or “simulation” does not tie into claim 1 as there is no simulation recited. It is not clear if this is an additional step to the steps recited in claim 1. If an additional step is required than the examiner recommend positively reciting the step in gerund form. For example, it is not clear if claim 6 requires receiving information or scenarios from one or more autonomous vehicle reporting databases and simulating realistic scenarios based on information from the one or more autonomous vehicle reporting databases. Claim 18-20 have similar limitations and are rejected for the same reason
Claim 5 recites “ wherein the sample paths of the independent adversary vehicle”. There is insufficient antecedent basis for this limitation in the claim. Claims 13 and 19 have similar recitations and are rejected for the same reasons.
Claim 5 recites “ wherein the sample paths of the independent adversary vehicle and the primary vehicle should not collide.” It is not clear the scope of the claim as written. The term “should” does not require that it does not collide, and thus the scope of the claim as written, means that it could or could not collide. The examiner believes Applicant intends to recite “wherein the sample paths of the independent adversary vehicle and the primary vehicle does not collide”. Claims 13 and 19 have similar recitations and are rejected for the same reasons.
Claim 6 recites “realistic scenarios are simulated”. There is insufficient antecedent basis for “realistic scenarios” in the claim. Claims 14 and 20 have similar recitations and are rejected for the same reasons.
The term “realistic” in claim 6 is a relative term which renders the claim indefinite. The term “realistic” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. It is not clear how to determine if a scenario is realistic or unrealistic. Claims 14 and 20 have similar recitations and are rejected for the same reasons.
Claim 6 recites “wherein realistic scenarios are simulated from one or more autonomous vehicle reporting databases”. It is not clear if claim 6 is further limiting one or more steps recited in claim 1 or if claim 6 is reciting additional steps of the method. There is insufficient antecedent basis for “realistic scenarios” in the claim. Claims 14 and 20 have similar recitations and are rejected for the same reasons.
Claim 7 recites “ the high-variance dataset”. There is insufficient antecedent basis for this limitation in the claim. Claim 15 has a similar recitation and is rejected for the same reasons.
Claim 8 recites “wherein one or more dissimilar collision and non-collision paths contribute to model robustness”. There is insufficient antecedent basis for this limitation in the claim. It is not clear if this claim is intending to refer back to the collision path and no collision path of claim 1. Claims 16 and 20 have similar recitations and are rejected for the same reasons.
Claim 9 recites “a no collision path of adversary vehicle” in line 21. Claim 9 previously recites an adversary vehicle in line 4. It is not clear if the adversary vehicle in line 21 is the same or different than that recited in line 4. The examiner recommends reciting “a no collision path of the adversary vehicle” or “a non-collision path of the adversary vehicle”, as appropriate. Claim 17 has a similar recitation and is rejected for the same reasons.
Claim 17 recites “between adversary and primary vehicle” in line 22. It is not clear if the adversary and primary vehicle are referring back to the adversary and primary vehicle recited in line 3 or the perturbation-adversary vehicle and primary vehicle recited in line 19.
Claim 17 recites “the Ne elite paths” in line 24 . There is insufficient antecedent basis for this limitation in the claim. The examiner notes that “the Ne elite path” was previously recited in the claim.
Claim 17 recites “a non-colliding path: and “a collision” in line 41. These limitations were previously recited in the claim at lines 24-26. It is not clear if the “non-colliding path” and “a collision” of line 41 are the same or different than those recited in lines 24-26.
Claim 17 recites “a machine learning model” in line 45. Claim 17 previously recited a machine learning model in line 27. It is not clear if the machine learning model of line 45 is the same or different than that recited in line 27.
Claim 17 recites “a high-variance data set” in line 44. Claim 17 previously recites “a high-variance data set” in line 26. Further, when reciting “the high-variance data set” it is not clear if the high-variance data set is referring back to the one introduced in line 44 or 26.
The examiner notes that claim 17 repeats the same claim language in two portions of the claim. Specifically claim 17 recites “ wherein the custom score function prioritizes a high priority sample path as a slight deviation from a non-colliding path that result in a collision in the Ne elite paths, wherein the custom score function prioritizes successively different colliding or non-colliding paths to generate a high-variance dataset, wherein the high-variance data set trains a machine learning model” in liens “ in lines 24-27 and recites “ wherein the custom score function prioritizes a high priority sample path as a slight deviation from a non-colliding path that result in a collision in the Ne elite paths; wherein the custom score function prioritizes successively different colliding or non-colliding paths to generate a high-variance dataset” It is not clear to the examiner if these are intended as separate steps. Further, repeating the claim language introduces clarity issues as address above, as it introduces terms more than once. The examiner has attempted to point them all out however, requests Applicants to carefully review the claims prior to the next response.
Claims 2-8 depend from claim 1 and are similarly rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, based on their dependency on claim 1.
Claims 10-16 depend from claim 9 and are similarly rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, based on their dependency on claim 9.
Claims 18-20 depend from claim 17 and are similarly rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, based on their dependency on claim 17.
The following is a quotation of 35 U.S.C. 112(d):
(d) REFERENCE IN DEPENDENT FORMS.—Subject to subsection (e), a claim in dependent form shall contain a reference to a claim previously set forth and then specify a further limitation of the subject matter claimed. A claim in dependent form shall be construed to incorporate by reference all the limitations of the claim to which it refers.
The following is a quotation of pre-AIA 35 U.S.C. 112, fourth paragraph:
Subject to the following paragraph [i.e., the fifth paragraph of pre-AIA 35 U.S.C. 112], a claim in dependent form shall contain a reference to a claim previously set forth and then specify a further limitation of the subject matter claimed. A claim in dependent form shall be construed to incorporate by reference all the limitations of the claim to which it refers.
Claims 5, 13 and 19 are rejected under 35 U.S.C. 112(d) or pre-AIA 35 U.S.C. 112, 4th paragraph, as being of improper dependent form for failing to further limit the subject matter of the claim upon which it depends, or for failing to include all the limitations of the claim upon which it depends. Claim 5 recites “ wherein the sample paths of the independent adversary vehicle and the primary vehicle should not collide.” The term “should” does not require that it does not collide, and thus the scope of the claim as written, means that it could or could not collide which are the only two options. Thus, the claim does not further limit the parent claim. Claims 13 and 19 have similar recitations and are rejected for the same reasons.
Applicant may cancel the claim(s), amend the claim(s) to place the claim(s) in proper dependent form, rewrite the claim(s) in independent form, or present a sufficient showing that the dependent claim(s) complies with the statutory requirements.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to an abstract idea without significantly more.
Following the 2019 Revised Patent Subject Matter Eligibility Guidance (84 Fed. Reg. 50-57 and MPEP § 2106, hereinafter 2019 Guidance), the claim(s) appear to recite at least one abstract idea, as explained in the Step 2A, Prong I analysis below. Furthermore, the judicial exception(s) does/do not appear to be integrated into a practical application as explained in the Step 2A, Prong II analysis below. Further still, the claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception(s) as explained in the Step 2B analysis below.
STEP 1:
Step 1, of the 2019 Guidance, first looks to whether the claimed invention is directed to a statutory category, namely a process, machine, manufactures, and compositions of matter.
Claim 1 is directed toward a method and is therefore eligible for further analysis.
Claim 9 is directed toward a system and is therefore eligible for further analysis.
Claim 17 is directed toward a method and is therefore eligible for further analysis.
STEP 2A, PRONG I:
Step 2A, prong I, of the 2019 Guidance, first looks to whether the claimed invention recites any judicial exceptions, including certain groupings of abstract ideas (i.e., mathematical concepts, certain methods of organizing human activities such as a fundamental economic practice, or mental processes).
Independent claim 6 includes limitations that recite an abstract idea (emphasized below) and will be used as a representative claim(s) for the remainder of the 101 rejection.
Claim 1 recites:
A method of training, verification, and advanced notice for an autonomous system, comprising:
by a machine learning classifier, wherein an adversary vehicle, and a primary vehicle are on the same path;
creating a path position probability transition matrix, wherein the path position probability transition matrix comprises an acceleration parameters array of one or more acceleration parameters for a potential position, wherein the path position probability transition matrix comprises equal probabilities for creating a samples array, evaluating a sample path, sorting a sample path, and using Ne elite path;
creating a samples array, wherein the sample array is one or more paths, wherein the one or more paths are a sequence of grid points of a two-dimensional grid, wherein the grid points comprise: a position and an acceleration, wherein the position is a sampled position from a probability distribution of the path probability transition matrix, and wherein the acceleration is a sampled acceleration from an acceleration probability distribution associated with the sampled position;
evaluating a sample path, wherein a custom score function asserts a potential position and a potential acceleration to identify a collision path of the perturbation-adversary and the primary vehicle, a no collision path of the adversary vehicle and the primary vehicle, a small position and acceleration distance between the adversary vehicle and the primary vehicle;
sorting the sample path, based at least in part a custom score of a custom score function and selecting a subset of the sample paths as Ne elite paths;
using the Ne elite paths to update the path probability transition matrix and the acceleration parameters array; and
repeating until the Ne elite paths stabilizes.
The examiner submits that the foregoing bolded limitation(s) constitute a “mental process” because under its broadest reasonable interpretation, the claim covers performance of the limitation in the human mind. Specifically, the “wherein an adversary vehicle, and a primary vehicle are on the same path” “creating a path position probability transition matrix, wherein the path position probability transition matrix comprises an acceleration parameters array of one or more acceleration parameters for a potential position, wherein the path position probability transition matrix comprises equal probabilities for creating a samples array, evaluating a sample path, sorting a sample path, and using Ne elite path;”, “creating a samples array, wherein the sample array is one or more paths, wherein the one or more paths are a sequence of grid points of a two-dimensional grid, wherein the grid points comprise: a position and an acceleration, wherein the position is a sampled position from a probability distribution of the path probability transition matrix, and wherein the acceleration is a sampled acceleration from an acceleration probability distribution associated with the sampled position;”, “evaluating a sample path, wherein a custom score function asserts a potential position and a potential acceleration to identify a collision path of the perturbation-adversary and the primary vehicle, a no collision path of the adversary vehicle and the primary vehicle, a small position and acceleration distance between the adversary vehicle and the primary vehicle”, “sorting the sample path, based at least in part a custom score of a custom score function and selecting a subset of the sample paths as Ne elite paths;”, “using the Ne elite paths to update the path probability transition matrix and the acceleration parameters array;” and “repeating until the Ne elite paths stabilizes.” steps encompass a human with a paper and pen assuming that the adversary vehicle and a primary vehicle are on the same path and utilizing mathematical equations to create a path position probability transition matrix by creating a grid, and determining a probability distribution based on sampled position and acceleration, evaluate the path by using a score system based on potential positions and accelerations to determine a possible collision, and sorting paths based on the score and selecting the most viable paths, and further using the most viable paths to update the transition matrix to repeat the process steps
STEP 2A, PRONG II:
Regarding Prong II of the Step 2A analysis in the 2019 PEG, the claims are to be analyzed to determine whether the claim, as a whole, integrates the abstract into a practical application. As noted in the 2019 PEG, it must be determined whether any additional elements in the claim beyond the abstract idea integrate the exception into a practical application in a manner that imposes a meaningful limit on the judicial exception. The courts have indicated that additional elements merely using a computer to implement an abstract idea, adding insignificant extra solution activity, or generally linking use of a judicial exception to a particular technological environment or field of use do not integrate a judicial exception into a “practical application”.
In the present case, the additional limitations beyond the above-noted abstract idea are as follows (where the underlined portions are the “additional limitations” while the bolded portions continue to represent the “abstract idea”):
Claim 1 recites:
A method of training, verification, and advanced notice for an autonomous system, comprising:
by a machine learning classifier, wherein an adversary vehicle, and a primary vehicle are on the same path;
creating a path position probability transition matrix, wherein the path position probability transition matrix comprises an acceleration parameters array of one or more acceleration parameters for a potential position, wherein the path position probability transition matrix comprises equal probabilities for creating a samples array, evaluating a sample path, sorting a sample path, and using Ne elite path;
creating a samples array, wherein the sample array is one or more paths, wherein the one or more paths are a sequence of grid points of a two-dimensional grid, wherein the grid points comprise: a position and an acceleration, wherein the position is a sampled position from a probability distribution of the path probability transition matrix, and wherein the acceleration is a sampled acceleration from an acceleration probability distribution associated with the sampled position;
evaluating a sample path, wherein a custom score function asserts a potential position and a potential acceleration to identify a collision path of the perturbation-adversary and the primary vehicle, a no collision path of the adversary vehicle and the primary vehicle, a small position and acceleration distance between the adversary vehicle and the primary vehicle;
sorting the sample path, based at least in part a custom score of a custom score function and selecting a subset of the sample paths as Ne elite paths;
using the Ne elite paths to update the path probability transition matrix and the acceleration parameters array; and
repeating until the Ne elite paths stabilizes.
For the following reason(s), the examiner submits that the above identified additional limitations do not integrate the above-noted abstract idea into a practical application:
Regarding the additional limitations of “by a machine learning classifier,” the examiner submits that these limitations merely using a computer to implement an abstract idea, adding insignificant extra solution activity, or generally linking use of a judicial exception to a particular technological environment or field of use and do not integrate a judicial exception into a “practical application”.
Specifically, the courts have held that merely reciting the works “apply it” (or an equivalent) with the judicial exception, or merely including or are more than mere instructions to implement an abstract idea on a computer, or merely using the computer as a tool to perform an abstract idea, does not integrate a judicial exception into a practical application. See MPEP 2106.05(f). The additional limitations of “by a machine learning classifier,” are recited at a high level of generality and simply describes using the computer as a tool to perform the abstract idea of “creating”, “evaluating”, “sorting” “using” and “repeating”. The additional limitations are no more than mere instructions to apply the exception using a general purpose computer (the examiner notes that eh machine learning classifier is an algorithm performed by a processor, see [0052-0055], [0119] of the instant application).
Thus, taken alone, the additional elements do not integrate the abstract idea into a practical application. Further, looking at the additional limitation(s) as an ordered combination or as a whole, the limitation(s) add nothing that is not already present when looking at the elements taken individually. For instance, there is no indication that the additional elements, when considered as a whole, reflect an improvement in the functioning of a computer or an improvement to another technology or technical field, apply or use the above-noted judicial exception to effect a particular treatment or prophylaxis for a disease or medical condition, implement/use the above-noted judicial exception with a particular machine or manufacture that is integral to the claim, effect a transformation or reduction of a particular article to a different state or thing, or apply or use the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment, such that the claim as a whole is not more than a drafting effort designed to monopolize the exception (MPEP § 2106.05). Accordingly, the additional limitation(s) do/does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
STEP 2B:
Regarding Step 2B of the Revised Guidance, the representative independent claim 1 does not include additional elements (considered both individually and as an ordered combination) that are sufficient to amount to significantly more than the judicial exception for the same reasons to those discussed above with respect to determining that the claim does not integrate the abstract idea into a practical application.
As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of “by a machine learning classifier,” amounts to nothing more than mere instructions to apply the exception using a generic computer or generic components (the examiner notes that the machine learning classifier is an algorithm performed by a processor, see [0052-0055], [0119] of the instant application). Mere instructions to apply an exception using a generic computer or generic components that are simply employed as a tool cannot provide an inventive concept.
Claims 9 and 17 have similar recitations to claim 1 and the analysis above with respect to claim 1 also applies to claims 9 and 17.
Dependent claim(s) 2-8, 10-16 and 18-20 do not recite any further limitations that cause the claim(s) to be patent eligible. Rather, the limitations of dependent claims are directed toward additional aspects of the judicial exception and/or well-understood, routine and conventional additional elements that do not integrate the judicial exception into a practical application. Specifically, the claims only recite limitations further defining the mental process (updating the path probability matrix of claim 2, 10, 17; prioritizing a high priority sample path of claims 3, 11, and 17; providing potential positions and acceleration of an independent adversary or vanilla adversary vehicle of claims 4, 12 and 18; adding random sample paths as recited in claim 7, 15) and insignificant extra-solution activity (from one or more autonomous vehicle reporting databases as recited in claims 6, 14, 20). These limitations are considered mental process steps and additional steps that amount to necessary data output. These additional elements fail to integrate the abstract idea into a practical application because they do not impose meaningful limits on the claimed invention. As such, the additional elements individually and in combination do not amount to significantly more than the abstract idea. Therefore, when considering the combination of elements and the claimed invention as a whole, claims 2-8, 10-16 and 18-20 are not patent eligible.
Accordingly, claims 1-20 are not patent eligible.
Claim Rejections - 35 USC § 102
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.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1-20 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by D. Drusinsky, et al. ("Machine-Learned Verification and Advance Notice Oracles for Autonomous Systems," in Computer, vol. 56, no. 7, pp. 121-130, July 2023, hereinafter “Drusinsky”).
Regarding claim 1, Drusinsky discloses a method of training, verification, and advanced notice for an autonomous system, comprising:
by a machine learning classifier, wherein an adversary vehicle, and a primary vehicle are on the same path (see at least Drusinsky Figure 1, and page 122-123 “We also apply the abovementioned 0-labeled, 1-labeled hybrid pairs as ML training datasets for the creation of ML-based VO classifiers (MLVOs). This contribution is elaborated in the section “Generating a High-Variance MLVO Dataset.” …The AV simulator we chose to use is Car Learning to Act (CARLA)1 due to its strong developer support and relatively open application programming interface….The primary vehicle, called the Ego vehicle, is assumed to be an ideal AV, with onboard AI that guides Ego to its preprogrammed destination while avoiding collisions with other vehicles. …In addition to the Ego, we model two additional vehicles that are not assumed to be autonomous or otherwise intelligent machines. The first vehicle, called the adversary (Adv), is a vehicle that potentially participates in a collision with Ego according to collision scenarios discovered by the CE falsifier, as described in the section “Hybrid Pair CE Search for AV Collision Scenarios.” In fact, the hybrid pair method discovers not only an Adv collision path but also a pair of very similar Adv paths, such that the vanilla path does not collide with Ego, while the perturbation path does. Additional details are available in the section “Hybrid Pair CE Search for AV Collision Scenarios.” See also page 125 and 128 regarding classifiers. For example the page 125 “When training an MLVO classifier, we classify vanilla paths as 0-labeled observations and perturbation paths as 1-labeled observations.” See also page 128 “In contrast, binary MLVO classifiers, being machine-learned objects, are evaluated using a confusion matrix (CM),2 depicted in Table 1.”);
creating a path position probability transition matrix, wherein the path position probability transition matrix comprises an acceleration parameters array of one or more acceleration parameters for a potential position, wherein the path position probability transition matrix comprises equal probabilities for creating a samples array, evaluating a sample path, sorting a sample path, and using Ne elite path (see at least Drusinsky Figure 1, wherein the ego vehicle is the primary vehicle and page 123-124 “Create the path probability transition matrix Mat. It is initialized with equal probabilities for all the possible next steps. Initialize an iteration counter t = 0. Let Ne = ρ × N, where the variable ρ is a small fraction such as 0.01” “Suppose now that we want to create a corresponding sequence of acceleration (or speed) values; that is, path[k], k = 0, 1, ... will consist of both the position and the acceleration in that step. To that end, we assume acceleration is normally distributed and therefore maintain normally distributed random variables. Specifically, that is, in addition to the n × m transition probability matrix Mat[i, j] used in Listing 1, we use an additional size n array called GP, which contains Gaussian parameter pairs (µ,σ2), one for each cell of the grid. Listing 1 is modified in steps 2, 4, and 5 accordingly, resulting in Listing 2. Listing 2: Pseudocode for the hybrid CE search algorithm 1. Execute step 1 of Listing 1. Create GP, an initial array of size n of normally distributed parameters. Initialize an iteration counter t = 0. 2. Create an array of N paths called samples. Each sample path is a sequence of points that encapsulates a grid position (denoted as pos) according to step 2 of Listing 1, as well as an acceleration value indicating the acceleration in pos. The acceleration value is stochastically generated by sampling from the probability distribution contained in GP[pos].” See also page 123 “The algorithm uses a transition probability matrix called Mat, whose size is n × m, where m is the number of possible maneuvers, or actions, that result in a certain path position at step k + 1 (path[k + 1]) given its position at step k (path[k]). When using a 2D grid representation of the underlying domain-of-discourse roadway (Figure 1), then n is the size of the grid (width times height as discussed in the section “Setting the Stage: AV Simulation”),
creating a samples array, wherein the sample array is one or more paths, wherein the one or more paths are a sequence of grid points of a two-dimensional grid, wherein the grid points comprise: a position and an acceleration, wherein the position is a sampled position from a probability distribution of the path probability transition matrix, and wherein the acceleration is a sampled acceleration from an acceleration probability distribution associated with the sampled position see at least Drusinsky Figure 1, and page 124 “Create an array of N paths called samples. Each sample path is a sequence of points created stochastically by sampling from the probability distribution contained in Mat, where N is typically 1–10 times larger than n. The source position is always path[0]; after that, path[j], j = 1, 2, ..., is the position attained as a result of one of m possible maneuvers; it is drawn probabilistically from the Mat[path[j]] m-categories distribution“ See further on page 124 “Suppose now that we want to create a corresponding sequence of acceleration (or speed) values; that is, path[k], k = 0, 1, ... will consist of both the position and the acceleration in that step. To that end, we assume acceleration is normally distributed and therefore maintain normally distributed random variables. Specifically, that is, in addition to the n × m transition probability matrix Mat[i, j] used in Listing 1, we use an additional size n array called GP, which contains Gaussian parameter pairs (µ,σ2), one for each cell of the grid. Listing 1 is modified in steps 2, 4, and 5 accordingly, resulting in Listing 2. Listing 2: Pseudocode for the hybrid CE search algorithm 1. Execute step 1 of Listing 1. Create GP, an initial array of size n of normally distributed parameters. Initialize an iteration counter t = 0. 2. Create an array of N paths called samples. Each sample path is a sequence of points that encapsulates a grid position (denoted as pos) according to step 2 of Listing 1, as well as an acceleration value indicating the acceleration in pos. The acceleration value is stochastically generated by sampling from the probability distribution contained in GP[pos].”),
evaluating a sample path, wherein a custom score function asserts a potential position and a potential acceleration to identify a collision path of the perturbation-adversary and the primary vehicle, a no collision path of the adversary vehicle and the primary vehicle, a small position and acceleration distance between the adversary vehicle and the primary vehicle (see at least Drusinsky page 124 “ Evaluate each path in the samples array using a custom score function, CE’s equivalent of a specification. Smaller numbered scores are considered to be superior to larger numbered scores. A negative score means the path conforms to all the rigid constraints, whereas a nonnegative score value means the path is inadmissible; for example, an Adv path that is expected to collide with Ego will be scored with a negative score if and only if that path indeed collides with Ego. The lower the score, the more the path conforms to the governing optimization and uncertainty constraints” See also further on page 124 “Evaluate each path in paths using the custom score function, which can now assert about both position and acceleration”);
sorting the sample path, based at least in part a custom score of a custom score function and selecting a subset of the sample paths as Ne elite paths (see at least Drusinsky page 124 “Sort the samples paths array by their ascending score. The first Ne paths of the sorted list (that is, the best paths according to their score) are called the elite set of sample paths. Terminate if samples[Ne ] has not changed in the last d iterations (for example, d = 4) and declare paths[0] as the best path” See also on page 124 “Sort paths by their score. Terminate as in step 4 of in Listing 1, and likewise declare samples[0] as the best path.”);
using the Ne elite paths to update the path probability transition matrix and the acceleration parameters array (see at least Drusinsky page 124 “;Use the Ne elite sample paths to update Mat using standard frequency counting, that is, Mat[i, j] = number of times any path of the elite set transitioned from location i to location j, divided by Ne” See also further on page 124 Use the Ne elite paths to update both Mat (see step 5 of Listing 1) and GP. For GP, use the Ne elite paths to update the Gaussian parameters pairs (µi,σ2) of GP[i], i in [0, n), as follows: µi = 1 NeΣNe k=0 accel[k, i] σi 2 = 1 NeΣNe k=0 (µi − accel[k, i])2 where accel[k, i] is the acceleration in cell i of the grid according to the kth elite path.”) and
repeating until the Ne elite paths stabilizes (see at least Drusinsky page 124, step 6, Under the listing 1 “Increment t and go to step 2.” And further on page 124 under listing 2 “Increment t and go to step 2.” Further see at least page 124 “Terminate if samples[Ne ] has not changed in the last d iterations (for example, d = 4) and declare paths[0] as the best path” See also page 127).
Regarding claim 2, Drusinsky teaches the method of claim 1, further comprising:
updating the path probability transition matrix by the number of times the Ne elite paths transitions from location i to location j, divided by N2; and
updating Gaussian parameters pairs of the acceleration parameters array as follows:
µi=1Ne∑K=0Neaccekk,i
σi2=1Ne∑k=0Ne(µi-accelk,i)2
wherein accel[k, i] is the acceleration in cell i of the two-dimensional grid according to a k elite path (see at least Drusinsky page 124 “;Use the Ne elite sample paths to update Mat using standard frequency counting, that is, Mat[i, j] = number of times any path of the elite set transitioned from location i to location j, divided by Ne” See also further on page 124 Use the Ne elite paths to update both Mat (see step 5 of Listing 1) and GP. For GP, use the Ne elite paths to update the Gaussian parameters pairs (µi,σ2) of GP[i], i in [0, n), as follows: µi = 1 NeΣNe k=0 accel[k, i] σi 2 = 1 NeΣNe k=0 (µi − accel[k, i])2 where accel[k, i] is the acceleration in cell i of the grid according to the kth elite path.”)
Regarding claim 3, Drusinsky teaches method of claim 1, wherein the custom score function prioritizes a high priority sample path as a slight deviation from a non-colliding path that result in a collision in the Ne elite paths
wherein the custom score function prioritizes successively different colliding or non-colliding paths to generate a high-variance dataset; and
wherein the high-variance data set trains a machine learning model (see at least Drusinsky Figure 2 and Figure 4. See additionally page 125 “Discover interesting colliding Adv paths. An example of a noninteresting Adv path is one where Adv deliberately swerves into Ego with no advance warning behavior; it is noninteresting because there is nothing Ego can do about it. Hence, we search for Adv paths that although they collide with Ego (that is, “bad”), they are also only slight deviations from an otherwise noncolliding (that is, “good”) path. In other words, we search for pairs of Adv paths, where the vanilla and perturbation paths differ in a slight modification of their parameters, such as visiting slightly different position(s) along their respective path or travelling with slightly different acceleration (or speed) values along some of their respective path positions. Figure 2 depicts an Ego along with such a pair. Note that the pair of scenarios is for the same vehicle (Adv); they are two slightly different behaviors of the same vehicle where the perturbation behavior results in a collision with Ego and the vanilla behavior does not.” See also page 127-128, Section “Generating a High-Variance and Explainable MLVO Datasets” “Recall how Adv path pairs generated by hybrid pair CE induce an abstract boundary between 0- and 1-labeled paths. In other words, a slight perturbation of the parameters of the 0-labeled (vanilla) path render it a 1-labeled path. A high-quality MLVO dataset, however, requires training data that are not necessarily on that borderline, for example, a vehicle traveling along a similar sequence of locations as the vanilla path, yet whose sequence of accelerations is sufficiently different from that of vanilla. To that end, we introduce rudimentary and variant path pairs. Rudimentary paths are those generated from distributions that are not iteratively updated by CE; in other words, they are akin to Monte Carlo generated paths. They represent naturalistic noncollision paths likely to be prevalent in randomly sampled real-world data. In contrast, variant vanilla (perturbation) paths are paths that do use the latest hybrid distribution generated by vanilla CE (perturbation CE) but are not the best vanilla (perturbation) path, that is, not the top of the elite set; rather, they are chosen randomly from within the elite set, noting that the entire elite set satisfies all rigid constraints (such as the vanilla path not colliding with Ego). In other words, variant paths are farther away from the abstract boundary line than their Adv path pair counterparts. All abovementioned types of paths used for MLVO training, as well as the abstract 0-labeled/1-labeled boundary, are depicted in Figure 4.” See also 122 and throughout article for machine learning model training )
Regarding claim 4, Drusinsky teaches method of claim 1, further comprising an independent adversary vehicle, wherein the independent adversary vehicle, the vanilla adversary vehicle, the adversary vehicle, and the primary vehicle are on the same path ; and
wherein the custom score function asserts a potential position and a potential acceleration to identify a no collision between the adversary vehicle and the independent adversary vehicle, and a small position and acceleration distance between the adversary vehicle and independent adversary vehicle (see at least Drusinsky Figure 3 and page 125-126 “We therefore enhance the hybrid pair approach to include yet another moving vehicle, called the IndAdv; IndAdv does not directly collide with Ego but contributes to Adv’s collision with Ego by encroaching into Adv’s lane, which in turn makes Adv move toward Ego to avoid a collision with IndAdv. This is depicted in Figure 3. Hence, the CE search system now contains three searches, two for Adv (vanilla and perturbation) and one for IndAdv. Listing 4 contains the constraints for IndAdv, while performing its CE (called IndAdv CE). The first three constraints are rigid constraints, while the remaining are optimization constraints” See also page 128 “The second vehicle, called the independent adversary (IndAdv), does not itself collide with Ego or with Adv; rather, it is added when needed to model a third-party vehicle that influences Adv to collide with Ego by encroaching on Adv’s path. Additional details are available in the section “Hybrid Pair CE Search With Additional Actors.”);
Regarding claim 5, Drusinsky teaches method of claim 4, wherein the sample paths of the independent adversary vehicle and the primary vehicle should not collide (see at least Drusinsky Figure 3 and page 125-126 “We therefore enhance the hybrid pair approach to include yet another moving vehicle, called the IndAdv; IndAdv does not directly collide with Ego but contributes to Adv’s collision with Ego by encroaching into Adv’s lane, which in turn makes Adv move toward Ego to avoid a collision with IndAdv. This is depicted in Figure 3. Hence, the CE search system now contains three searches, two for Adv (vanilla and perturbation) and one for IndAdv. Listing 4 contains the constraints for IndAdv, while performing its CE (called IndAdv CE). The first three constraints are rigid constraints, while the remaining are optimization constraints…The IndAdv path should not collide with Ego.” page 128 “The second vehicle, called the independent adversary (IndAdv), does not itself collide with Ego or with Adv; rather, it is added when needed to model a third-party vehicle that influences Adv to collide with Ego by encroaching on Adv’s path. Additional details are available in the section “Hybrid Pair CE Search With Additional Actors.”);.
Regarding claim 6, Drusinsky teaches method of claim 1, wherein realistic scenarios are simulated from one or more autonomous vehicle reporting databases (see at least Drusinsky page 127-128 “To that end, we introduce rudimentary and variant path pairs. Rudimentary paths are those generated from distributions that are not iteratively updated by CE; in other words, they are akin to Monte Carlo generated paths. They represent naturalistic noncollision paths likely to be prevalent in randomly sampled real-world data.” See further on page 128 “We use the California Department of Motor Vehicles (DMV) disengagement report database. In 2014, the California DMV began requiring companies testing AVs on California public roads to submit disengagement reports. These reports record instances where the vehicle’s autonomous control system is disengaged, either by the vehicle or by the human backup driver, and the California DMV publishes on an annual basis the disengagement reports (as well as crash reports) from all companies testing AVs. Hence, consider the California DMV report line item (denoted DMV1)”
Regarding claim 7, Drusinsky teaches method of claim 1, wherein one or more random sample paths are added to the high-variance dataset (see Drusinsky page 127-128, Section “Generating a High-Variance and Explainable MLVO Datasets” “Recall how Adv path pairs generated by hybrid pair CE induce an abstract boundary between 0- and 1-labeled paths. In other words, a slight perturbation of the parameters of the 0-labeled (vanilla) path render it a 1-labeled path. A high-quality MLVO dataset, however, requires training data that are not necessarily on that borderline, for example, a vehicle traveling along a similar sequence of locations as the vanilla path, yet whose sequence of accelerations is sufficiently different from that of vanilla. To that end, we introduce rudimentary and variant path pairs. Rudimentary paths are those generated from distributions that are not iteratively updated by CE; in other words, they are akin to Monte Carlo generated paths. They represent naturalistic noncollision paths likely to be prevalent in randomly sampled real-world data. In contrast, variant vanilla (perturbation) paths are paths that do use the latest hybrid distribution generated by vanilla CE (perturbation CE) but are not the best vanilla (perturbation) path, that is, not the top of the elite set; rather, they are chosen randomly from within the elite set, noting that the entire elite set satisfies all rigid constraints (such as the vanilla path not colliding with Ego). In other words, variant paths are farther away from the abstract boundary line than their Adv path pair counterparts. All abovementioned types of paths used for MLVO training, as well as the abstract 0-labeled/1-labeled boundary, are depicted in Figure 4.” and
wherein the one or more random sample paths represent one or more naturalistic paths (see at least Drusinsky page 127-128 “To that end, we introduce rudimentary and variant path pairs. Rudimentary paths are those generated from distributions that are not iteratively updated by CE; in other words, they are akin to Monte Carlo generated paths. They represent naturalistic noncollision paths likely to be prevalent in randomly sampled real-world data.” )
Regarding claim 8, Drusinsky teaches method of claim 1, wherein one or more dissimilar collision and non-collision paths contribute to model robustness (see at least Drusinsky page 123 “In fact, the hybrid pair method discovers not only an Adv collision path but also a pair of very similar Adv paths, such that the vanilla path does not collide with Ego, while the perturbation path does. Additional details are available in the section “Hybrid Pair CE Search for AV Collision Scenarios.” See also page 124-125 section “Hybrid Pair CE Search for AV Collision Scenarios.” See also page 127-128 “To that end, we introduce rudimentary and variant path pairs. Rudimentary paths are those generated from distributions that are not iteratively updated by CE; in other words, they are akin to Monte Carlo generated paths. They represent naturalistic noncollision paths likely to be prevalent in randomly sampled real-world data.” ).
Claim 9 is rejected under the same rationale, mutatis mutandis, as claim 1, above.
Claim 10 is rejected under the same rationale, mutatis mutandis, as claim 2, above.
Claim 11 is rejected under the same rationale, mutatis mutandis, as claim 3, above.
Claim 12 is rejected under the same rationale, mutatis mutandis, as claim 4, above.
Claim 13 is rejected under the same rationale, mutatis mutandis, as claim 5, above.
Claim 14 is rejected under the same rationale, mutatis mutandis, as claim 6, above.
Claim 15 is rejected under the same rationale, mutatis mutandis, as claim 7, above.
Claim 16 is rejected under the same rationale, mutatis mutandis, as claim 8, above.
Regarding claim 17, Drusinsky discloses a Hybrid-Pair-Cross Entropy method of training, verification, and advanced notice for an autonomous system, the method comprising:
by an adversary vehicle and a primary vehicle, wherein the adversary vehicle, and the primary vehicle are on the same path (see at least Drusinsky Figure 1, and page 122-123 “We also apply the abovementioned 0-labeled, 1-labeled hybrid pairs as ML training datasets for the creation of ML-based VO classifiers (MLVOs). This contribution is elaborated in the section “Generating a High-Variance MLVO Dataset.” …The AV simulator we chose to use is Car Learning to Act (CARLA)1 due to its strong developer support and relatively open application programming interface….The primary vehicle, called the Ego vehicle, is assumed to be an ideal AV, with onboard AI that guides Ego to its preprogrammed destination while avoiding collisions with other vehicles. …In addition to the Ego, we model two additional vehicles that are not assumed to be autonomous or otherwise intelligent machines. The first vehicle, called the adversary (Adv), is a vehicle that potentially participates in a collision with Ego according to collision scenarios discovered by the CE falsifier, as described in the section “Hybrid Pair CE Search for AV Collision Scenarios.” In fact, the hybrid pair method discovers not only an Adv collision path but also a pair of very similar Adv paths, such that the vanilla path does not collide with Ego, while the perturbation path does. Additional details are available in the section “Hybrid Pair CE Search for AV Collision Scenarios.” See also page 125 and 128 regarding classifiers. For example the page 125 “When training an MLVO classifier, we classify vanilla paths as 0-labeled observations and perturbation paths as 1-labeled observations.” See also page 128 “In contrast, binary MLVO classifiers, being machine-learned objects, are evaluated using a confusion matrix (CM),2 depicted in Table 1.”);
creating a path position probability transition matrix, wherein the path position probability transition matrix comprises an acceleration parameters array of one or more acceleration parameters for a potential position, wherein the path position probability transition matrix comprises equal probabilities for creating a samples array, evaluating a sample path, sorting a sample path, and using Ne elite path (see at least Drusinsky Figure 1, wherein the ego vehicle is the primary vehicle and page 123-124 “Create the path probability transition matrix Mat. It is initialized with equal probabilities for all the possible next steps. Initialize an iteration counter t = 0. Let Ne = ρ × N, where the variable ρ is a small fraction such as 0.01” “Suppose now that we want to create a corresponding sequence of acceleration (or speed) values; that is, path[k], k = 0, 1, ... will consist of both the position and the acceleration in that step. To that end, we assume acceleration is normally distributed and therefore maintain normally distributed random variables. Specifically, that is, in addition to the n × m transition probability matrix Mat[i, j] used in Listing 1, we use an additional size n array called GP, which contains Gaussian parameter pairs (µ,σ2), one for each cell of the grid. Listing 1 is modified in steps 2, 4, and 5 accordingly, resulting in Listing 2. Listing 2: Pseudocode for the hybrid CE search algorithm 1. Execute step 1 of Listing 1. Create GP, an initial array of size n of normally distributed parameters. Initialize an iteration counter t = 0. 2. Create an array of N paths called samples. Each sample path is a sequence of points that encapsulates a grid position (denoted as pos) according to step 2 of Listing 1, as well as an acceleration value indicating the acceleration in pos. The acceleration value is stochastically generated by sampling from the probability distribution contained in GP[pos].” See also page 123 “The algorithm uses a transition probability matrix called Mat, whose size is n × m, where m is the number of possible maneuvers, or actions, that result in a certain path position at step k + 1 (path[k + 1]) given its position at step k (path[k]). When using a 2D grid representation of the underlying domain-of-discourse roadway (Figure 1), then n is the size of the grid (width times height as discussed in the section “Setting the Stage: AV Simulation”);
creating a samples array, wherein the sample array is one or more paths, wherein the one or more paths are a sequence of grid points of a two-dimensional grid, wherein the grid points comprise: a position and an acceleration, wherein the position is a sampled position from a probability distribution of the path probability transition matrix, and wherein the acceleration is a sampled acceleration from an acceleration probability distribution associated with the sampled position (see at least Drusinsky Figure 1, and page 124 “Create an array of N paths called samples. Each sample path is a sequence of points created stochastically by sampling from the probability distribution contained in Mat, where N is typically 1–10 times larger than n. The source position is always path[0]; after that, path[j], j = 1, 2, ..., is the position attained as a result of one of m possible maneuvers; it is drawn probabilistically from the Mat[path[j]] m-categories distribution“ See further on page 124 “Suppose now that we want to create a corresponding sequence of acceleration (or speed) values; that is, path[k], k = 0, 1, ... will consist of both the position and the acceleration in that step. To that end, we assume acceleration is normally distributed and therefore maintain normally distributed random variables. Specifically, that is, in addition to the n × m transition probability matrix Mat[i, j] used in Listing 1, we use an additional size n array called GP, which contains Gaussian parameter pairs (µ,σ2), one for each cell of the grid. Listing 1 is modified in steps 2, 4, and 5 accordingly, resulting in Listing 2. Listing 2: Pseudocode for the hybrid CE search algorithm 1. Execute step 1 of Listing 1. Create GP, an initial array of size n of normally distributed parameters. Initialize an iteration counter t = 0. 2. Create an array of N paths called samples. Each sample path is a sequence of points that encapsulates a grid position (denoted as pos) according to step 2 of Listing 1, as well as an acceleration value indicating the acceleration in pos. The acceleration value is stochastically generated by sampling from the probability distribution contained in GP[pos].”);
evaluating a sample path, wherein a custom score function asserts a potential position and a potential acceleration to identify a collision path of the perturbation-adversary and the primary vehicle , a no collision path of adversary vehicle and the primary vehicle , a small position and acceleration distance between adversary and primary vehicle, wherein the custom score function prioritizes a high priority sample path as a slight deviation from a non-colliding path that result in a collision in the Ne elite paths, wherein the custom score function prioritizes successively different colliding or non-colliding paths to generate a high-variance dataset, wherein the high-variance data set trains a machine learning model (see at least Drusinsky page 124 “ Evaluate each path in the samples array using a custom score function, CE’s equivalent of a specification. Smaller numbered scores are considered to be superior to larger numbered scores. A negative score means the path conforms to all the rigid constraints, whereas a nonnegative score value means the path is inadmissible; for example, an Adv path that is expected to collide with Ego will be scored with a negative score if and only if that path indeed collides with Ego. The lower the score, the more the path conforms to the governing optimization and uncertainty constraints” See also further on page 124 “Evaluate each path in paths using the custom score function, which can now assert about both position and acceleration” See also Drusinsky Figure 2 and Figure 4. See additionally page 125 “Discover interesting colliding Adv paths. An example of a noninteresting Adv path is one where Adv deliberately swerves into Ego with no advance warning behavior; it is noninteresting because there is nothing Ego can do about it. Hence, we search for Adv paths that although they collide with Ego (that is, “bad”), they are also only slight deviations from an otherwise noncolliding (that is, “good”) path. In other words, we search for pairs of Adv paths, where the vanilla and perturbation paths differ in a slight modification of their parameters, such as visiting slightly different position(s) along their respective path or travelling with slightly different acceleration (or speed) values along some of their respective path positions. Figure 2 depicts an Ego along with such a pair. Note that the pair of scenarios is for the same vehicle (Adv); they are two slightly different behaviors of the same vehicle where the perturbation behavior results in a collision with Ego and the vanilla behavior does not.” See also page 127-128, Section “Generating a High-Variance and Explainable MLVO Datasets” “Recall how Adv path pairs generated by hybrid pair CE induce an abstract boundary between 0- and 1-labeled paths. In other words, a slight perturbation of the parameters of the 0-labeled (vanilla) path render it a 1-labeled path. A high-quality MLVO dataset, however, requires training data that are not necessarily on that borderline, for example, a vehicle traveling along a similar sequence of locations as the vanilla path, yet whose sequence of accelerations is sufficiently different from that of vanilla. To that end, we introduce rudimentary and variant path pairs. Rudimentary paths are those generated from distributions that are not iteratively updated by CE; in other words, they are akin to Monte Carlo generated paths. They represent naturalistic noncollision paths likely to be prevalent in randomly sampled real-world data. In contrast, variant vanilla (perturbation) paths are paths that do use the latest hybrid distribution generated by vanilla CE (perturbation CE) but are not the best vanilla (perturbation) path, that is, not the top of the elite set; rather, they are chosen randomly from within the elite set, noting that the entire elite set satisfies all rigid constraints (such as the vanilla path not colliding with Ego). In other words, variant paths are farther away from the abstract boundary line than their Adv path pair counterparts. All abovementioned types of paths used for MLVO training, as well as the abstract 0-labeled/1-labeled boundary, are depicted in Figure 4.” See also 122 and throughout article for machine learning model training );
sorting the sample path based on a corresponding score from the custom score function and selecting a subset of the sample paths as Ne elite paths (see at least Drusinsky page 124 “Sort the samples paths array by their ascending score. The first Ne paths of the sorted list (that is, the best paths according to their score) are called the elite set of sample paths. Terminate if samples[Ne ] has not changed in the last d iterations (for example, d = 4) and declare paths[0] as the best path” See also on page 124 “Sort paths by their score. Terminate as in step 4 of in Listing 1, and likewise declare samples[0] as the best path.”);
using the Ne elite paths to update the path probability transition matrix and the acceleration parameters array (see at least Drusinsky page 124 “;Use the Ne elite sample paths to update Mat using standard frequency counting, that is, Mat[i, j] = number of times any path of the elite set transitioned from location i to location j, divided by Ne” See also further on page 124 Use the Ne elite paths to update both Mat (see step 5 of Listing 1) and GP. For GP, use the Ne elite paths to update the Gaussian parameters pairs (µi,σ2) of GP[i], i in [0, n), as follows: µi = 1 NeΣNe k=0 accel[k, i] σi 2 = 1 NeΣNe k=0 (µi − accel[k, i])2 where accel[k, i] is the acceleration in cell i of the grid according to the kth elite path.”) and
repeating until the Ne elite paths stabilizes for a predetermined number of iterations (see at least Drusinsky page 124, step 6, Under the listing 1 “Increment t and go to step 2.” And further on page 124 under listing 2 “Increment t and go to step 2.” Further see at least page 124 “Terminate if samples[Ne ] has not changed in the last d iterations (for example, d = 4) and declare paths[0] as the best path” See also page 127).
updating the path probability transition matrix by the number of times the Ne elite paths transitions from location i to location j, divided by N2;
updating Gaussian parameters pairs of the acceleration parameters array as follows:
µi=1Ne∑K=0Neaccekk,i
σi2=1Ne∑k=0Ne(µi-accelk,i)2
wherein accel[k, i] is the acceleration in cell i of the two-dimensional grid according to a k elite path (see at least Drusinsky page 124 “;Use the Ne elite sample paths to update Mat using standard frequency counting, that is, Mat[i, j] = number of times any path of the elite set transitioned from location i to location j, divided by Ne” See also further on page 124 Use the Ne elite paths to update both Mat (see step 5 of Listing 1) and GP. For GP, use the Ne elite paths to update the Gaussian parameters pairs (µi,σ2) of GP[i], i in [0, n), as follows: µi = 1 NeΣNe k=0 accel[k, i] σi 2 = 1 NeΣNe k=0 (µi − accel[k, i])2 where accel[k, i] is the acceleration in cell i of the grid according to the kth elite path.”);
wherein the custom score function prioritizes a high priority sample path as a slight deviation from a non-colliding path that result in a collision in the Ne elite paths;
wherein the custom score function prioritizes successively different colliding or non-colliding paths to generate a high-variance dataset; and
wherein the high-variance data set trains a machine learning model (see at least Drusinsky Figure 2 and Figure 4. See additionally page 125 “Discover interesting colliding Adv paths. An example of a noninteresting Adv path is one where Adv deliberately swerves into Ego with no advance warning behavior; it is noninteresting because there is nothing Ego can do about it. Hence, we search for Adv paths that although they collide with Ego (that is, “bad”), they are also only slight deviations from an otherwise noncolliding (that is, “good”) path. In other words, we search for pairs of Adv paths, where the vanilla and perturbation paths differ in a slight modification of their parameters, such as visiting slightly different position(s) along their respective path or travelling with slightly different acceleration (or speed) values along some of their respective path positions. Figure 2 depicts an Ego along with such a pair. Note that the pair of scenarios is for the same vehicle (Adv); they are two slightly different behaviors of the same vehicle where the perturbation behavior results in a collision with Ego and the vanilla behavior does not.” See also page 127-128, Section “Generating a High-Variance and Explainable MLVO Datasets” “Recall how Adv path pairs generated by hybrid pair CE induce an abstract boundary between 0- and 1-labeled paths. In other words, a slight perturbation of the parameters of the 0-labeled (vanilla) path render it a 1-labeled path. A high-quality MLVO dataset, however, requires training data that are not necessarily on that borderline, for example, a vehicle traveling along a similar sequence of locations as the vanilla path, yet whose sequence of accelerations is sufficiently different from that of vanilla. To that end, we introduce rudimentary and variant path pairs. Rudimentary paths are those generated from distributions that are not iteratively updated by CE; in other words, they are akin to Monte Carlo generated paths. They represent naturalistic noncollision paths likely to be prevalent in randomly sampled real-world data. In contrast, variant vanilla (perturbation) paths are paths that do use the latest hybrid distribution generated by vanilla CE (perturbation CE) but are not the best vanilla (perturbation) path, that is, not the top of the elite set; rather, they are chosen randomly from within the elite set, noting that the entire elite set satisfies all rigid constraints (such as the vanilla path not colliding with Ego). In other words, variant paths are farther away from the abstract boundary line than their Adv path pair counterparts. All abovementioned types of paths used for MLVO training, as well as the abstract 0-labeled/1-labeled boundary, are depicted in Figure 4.” See also 122 and throughout article for machine learning model training ).
Claim 18 is rejected under the same rationale, mutatis mutandis, as claim 4, above.
Claim 19 is rejected under the same rationale, mutatis mutandis, as claim 5, above.
Claim 20 is rejected under the same rationale, mutatis mutandis, as claim 8, above.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
D. Drusinsky and J. B. Michael, "Multiagent Pathfinding Under Rigid, Optimization, and Uncertainty Constraints," in Computer, vol. 54, no. 7, pp. 111-118, July 2021, is cited for showing an algorithm of cross-entropy optimization relevant to pathfinding in autonomous vehicles.
Drusinsky US Pub. No. 2023/0179512 is cited for disclosing multiagent pathfinding including a cross-entropy method which provides for sampling from a complex probability distribution relevant to pathfinding in autonomous vehicles.
Dupray et al US Pub. No. 2017/0069214 is cited for showing determining the probability of collision between at least two vehicles based on the probability of the positions and acceleration (see at least Dupray Figure 3 and 4A-4D [0233]) .
US-12258008-B1 to Kurutach and US-12485882-B1 to Wang are cited for showing collision path prediction.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JENNIFER M. ANDA whose telephone number is (571)272-5042. The examiner can normally be reached Monday-Friday 8:30 am-5pm MST.
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, Aniss Chad can be reached on (571)270-3832. 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.
/JENNIFER M ANDA/Examiner, Art Unit 3662