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 .
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 (i.e., changing from AIA to pre-AIA ) 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 text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claim(s) 1 and 10 is/are rejected under 35 U.S.C. 103 as being unpatentable over Agam US 20230205533 A1 (hereinafter Agam) in view of Shuai et al. CN 114154231 A (hereinafter Shuai; a translated copy has been provided which the examiner relies upon).
Regarding claim 1,
Agam teaches
An incremental neural network model inference method for updating a navigation state estimate of a vehicle, the method comprising:
providing a neural network model to a disk storage (¶ 0076-0077 disclose a processing unit comprising a processor and disk drive memory wherein the memory stores software to perform the method including trained neural networks) of a computer onboard the vehicle (Figure 2A shows the processing unit 110 is aboard the vehicle), wherein the computer is configured to execute the neural network model onboard the vehicle (¶ 0076 discloses software is executed by the processor), and the neural network model comprises a plurality of layers (see at least claim 7 “the neural network comprises a plurality of layers”);
reading parameters and functions from a layer of the plurality of layers into a memory of the computer (¶ 0006 discloses caching neural network coefficients related to a particular layer of the neural network used by neural network processors in iterative processing of neural network operations);
transforming a navigation state estimate with the parameters and functions (¶ 0006 discloses neural network processors perform neural network operations wherein cached coefficients are used by the processors; ¶ 0393 discloses network operations are related to functions of modules including velocity, acceleration, and navigation response modules indicating results include navigational states; see at least ¶ 0157 regarding an example of performing processing to construct a vehicle path and ¶ 0256 which performs processing to obtain a distance to a landmark in the environment and ¶ 0397 which details final results comprise various navigational states) and saving a transformed navigation state estimate to the memory (¶ 0006 discloses storing intermediate results in memory);
determining whether the layer of the neural network model currently in memory is a last layer of the plurality of layers (¶ 0380 discloses determining that there is no other layer to be selected after performing operations);
removing the parameters and functions from the memory (¶ 0006 discloses that coefficients used for operations are cached up to a first duration implying removal of cached coefficients after the first duration and ¶ 0360 details that coefficients are only stored for a single iteration; this implies a deletion, removal, or discarding of the cached coefficients occurs when the duration/iteration expires), while the transformed navigation state estimate remains in memory (Figure 28 2805, for example, discloses that intermediate results related to a particular layer are stored in memory before the next iteration; see also ¶ 0382);
incrementing to a next layer of the neural network model (¶ 0006 discloses the method is performed iteratively; see also claim 7 “repeating executing, storing, and retrieving for each of the plurality of layers” and ¶ 0380 wherein an example of incrementing from a first layer to a second layer is given);
repeating the steps of reading, transforming, and removing for the next layer of the neural network model (¶ 0006 discloses the method is performed iteratively; see also claim 7 “repeating executing, storing, and retrieving for each of the plurality of layers” and ¶ 0380 which discloses repeating operations related to the entire neural network); and
upon determining that the layer of the neural network model layer is determined to be the last layer, outputting a navigation command (¶ 0380 discloses that when there is no other layer to select, the processing ends; ¶ 0006 discloses final result of processing operations can be obtained and a navigational action can be performed in reaction to the final result); and
executing the navigation command for a current target epoch (¶ 0398-0399 discloses the navigational command is a command implemented to adjust or maintain a current movement such as a current heading).
Agam does not teach
transforming a navigation state estimate with the parameters and functions by performing a linear transformation on the navigation state estimate; and
upon determining that the layer of the neural network model layer is not the last layer, performing a nonlinear transformation on the transformed navigation state estimate;
Shuai teaches that
transforming a navigation state estimate with the parameters and functions by performing a linear transformation on the navigation state estimate (¶ 0022 discloses performing a linear transformation of input data in select layers of a neural network); and
upon determining that the layer of the neural network model layer is not the last layer, performing a nonlinear transformation on the transformed navigation state estimate (¶ 0022 discloses that specific layers, which are not the final layer, perform a non-linear calculation on the already transformed input).
There is only a finite list of options for the transformation functions: non-linear functions, linear functions, or a combination thereof. Therefore, it would have been prima facie obvious to one of ordinary skill in the art at the time of filing to try the teachings of Shuai and incorporate it into the teachings of Agam since there is a finite number of identified, predictable potential solutions (i.e. function types) to the recognized need (transforming input data) and one of ordinary skill in the art could have pursued the known potential solutions with a reasonable expectation of success. Furthermore, it would have been prima facie obvious to one of ordinary skill in the art at the time of filing to have modified Agam to incorporate the teachings of Shuai such that non-linear calculation does not occur on the final layer as taught by Shuai. This modification would be made with a reasonable expectation of success to reduce processing time for the final layer.
Regarding claim 10, Agam teaches all of claim 1 as detailed above.
Agam further teaches that
the step of transforming the navigation state estimate with the parameters and functions is performed for one layer of the neural network model at a time (¶ 0006 discloses the method is performed iteratively; see also claim 7 “repeating executing, storing, and retrieving for each of the plurality of layers” and ¶ 0380 which discloses repeating operations related to the entire neural network one selected layer at a time).
Claim(s) 7 is/are rejected under 35 U.S.C. 103 as being unpatentable over Agam in view of Shuai as applied to claim 1 above, and further in view of Xia et al. CN 114383617 A (hereinafter Xia; a translated copy has been provided which the examiner relies upon).
Regarding claim 7, Agam teaches all of claim 1 as detailed above.
Agam teaches that
the neural network model is previously trained (at least ¶ 0077 discloses the neural network is a trained system)
Agam does not explicitly teach that
the parameters and functions are previously obtained by training the neural network model on a set of possible trajectories for a maneuver of the vehicle.
Xia teaches that
the parameters and functions are previously obtained by training the neural network model on a set of possible trajectories for a maneuver of the vehicle (¶ 0009-0012 discloses a neural network is trained using historic vehicle trajectory data to adjust weights of nodes; examiner understands historic trajectories as equivalent to “possible trajectories” as they are ground truth data that was previously performed).
It would have been prima facie obvious to one of ordinary skill in the art at the time of filing to have modified Agam to incorporate the teachings of Xia such that the neural network comprising coefficients related to layers utilized for operations of Agam can be node weights, trained on historic trajectory data as taught by Xia. This modification would be made with a reasonable expectation of success to improve accuracy and robustness of the neural network model by training on real, historic data.
Claim(s) 9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Agam in view of Shuai as applied to claim 1 above, and further in view of Rohani et al. US 10796204 B2 (hereinafter Rohani).
Regarding claim 9, Agam teaches all of claim 1 as detailed above.
Agam does not teach that
the navigation state estimate is progressively transformed upon repeating steps for the next layer of the neural network model.
Rohani teaches that
the navigation state estimate is progressively transformed upon repeating steps for the next layer of the neural network model (Figure 10 shows that nodes are connected such that the output of one node is fed into every other node in the next layer; see also column 21 lines 29-67).
It would have been prima facie obvious to one of ordinary skill in the art at the time of filing to have modified Agam to incorporate the teachings of Rohani such that the neural network layers are intricately connected wherein the results of iterations of one layer of a neural network of Agam can be fed into a following layer as taught by Rohani. This modification would be made with a reasonable expectation of success to improve accuracy of results by iteratively adjusting results utilizing differing coefficients present in each layer.
Response to Amendment
Claim amendments filed 7/15/2026 have been received and fully considered and overcome the 112(b) and 101 rejections of record detailed in the Office Action dated 4/15/2026. These/this rejections have/has been withdrawn.
Response to Arguments
Applicant's arguments, see pages 8-10, filed 7/15/2026 have been fully considered but they are not persuasive. Applicant argues that the amended claims, as a whole, overcome the prior art since “amended claim 1 includes (1) a linear transformation on the navigation state estimate, (2) a determination of whether that layer is the last layer, and (3) only if it is not the last layer, an additional nonlinear transformation” which the prior of record, as alleged by the applicant, does not include. As indicated by the applicant, Shuai was utilized in the Office Action dated 4/15/2026 to fulfill these limitations and cure the deficiencies of Agam in regards to these limitations. Applicant argues that “the second fully-connected layer's linear output is immediately followed by a nonlinear activation-function calculation before the scaling layer outputs the estimated state quantity. In other words, in Shuai, the last linear-transformation stage is also followed by a nonlinear transformation” and that “The Examiner's rationale… ignores Shuai's actual disclosed order of operations”. Examiner respectfully asserts that the applicant is misunderstanding the purpose of the examiner’s cited section of Shuai (¶ 0022). Applicant appears to mistakenly understand the first layer as a layer that is not the final layer and the second layer as the final layer. Examiner respectfully indicates that the last layer of Shuai as detailed in ¶ 0022 is the scale layer. This can be seen in ¶ 0022 and better visually seen in at least Figure 5 which is machine translated with Google Image Translate below:
PNG
media_image1.png
200
400
media_image1.png
Greyscale
As shown, the scale layer (translated by Google from “缩放层” to “zoom layer” which, Google text translate indicates can also translate to “scaling layer”) is the final layer of the neural network. Scaling is a well-known linear function within the art. However, examiner understands that name alone, due to machine translation issues, may not be sufficient to provide full proof that the scale layer performs scaling. Thus, examiner asserts that ¶ 0078 gives more clarity about the function of the scale layer:
“The main function of the scaling layer is to scale the output value of the second activation function layer according to the actual possible value range of the state quantity to be estimated that is finally output by the second activation function layer, and its main parameters include the scale gain and Bias. The scaling layer of the strategy function neural network A1 outputs the estimated value of the second type of state quantity, and the scaling layer of the value function neural network A2 outputs the estimated effect evaluation value.”
As scaling is a well-known linear function within the art, and Shuai does not teach the additional performance of a non-linear transformation within the scale layer, it is clear that Shuai factually does not teach that the last layer (the scale layer) performs a non-linear function. Therefore, the applicant’s arguments are not persuasive and the 103 rejection in light of the prior art of record is maintained.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Ashley Tiffany Schoech whose telephone number is (571)272-2937. The examiner can normally be reached 4:45 am - 3:15 pm PT Monday - Thursday.
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, Erin Piateski can be reached at 571-270-7429. 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.
/A.T.S./Examiner, Art Unit 3669
/Erin M Piateski/Supervisory Patent Examiner, Art Unit 3669
Prosecution Insights