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 .
Response to Arguments
Applicant’s arguments with respect to claim(s) 1-20 have been considered but are moot because new grounds of rejection are made in view of Ha (US012198245B2).
Applicant argues Ambrus (US 20220301203A1) does not disclose loss function with distribution loss term that regularizes an output of model to produce realistic scenarios (p. 8-9).
In reply, the Examiner points out that Ambrus describes “the prediction system 170 may rank depth points that are unmasked according to a radial penalty using a Gaussian distribution for loss calculations…As such, the prediction system 170 may emphasize object centers in unmasked areas while gradually penalizing localization errors according to a distribution for keypoint processing. For instance, the prediction system 170 may penalize an object center off by one pixel less than other pixel errors for smoother estimation during inference” [0032]. Thus, it penalizes localization errors according to the distribution loss term, and thus this is a loss function with a distribution loss term that regularizes an output of the model to produce realistic scenarios [0032].
Applicant argues that none of the cited references teaches using a distribution loss to guide volumetric sampling density around a target region during rendering of a three-dimensional scene (p. 10).
In reply, the Examiner points out that new grounds of rejection are made in view of Ha to teach wherein the distribution loss term is configured to guide the model to increase sample density around a target region.
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 following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1, 3, 8-11, 13, and 18-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ambrus (US 20220301203A1), Ha (US012198245B2), and Ivanovic (US 20230391365A1).
As per Claim 1, Ambrus teaches a computer-implemented method, comprising: training a model for rendering a three-dimensional volume using a loss function that includes a depth loss term and a distribution loss term that regularizes an output of the model to produce realistic scenarios (reconstruct the structure of a scene with high accuracy, [0003], training to improve monocular depth estimates by adapting depth losses, three-dimensional detection network to determine the position, orientation, of objects in a 3D scene, [0027], prediction system 170 may rank depth points according to a radial penalty using a Gaussian distribution for loss calculations, prediction system 170 may also train the depth model by ranking depth points, [0032]). Ambrus describes “the prediction system 170 may rank depth points that are unmasked according to a radial penalty using a Gaussian distribution for loss calculations…As such, the prediction system 170 may emphasize object centers in unmasked areas while gradually penalizing localization errors according to a distribution for keypoint processing. For instance, the prediction system 170 may penalize an object center off by one pixel less than other pixel errors for smoother estimation during inference” [0032]. Thus, it penalizes localization errors according to the distribution loss term, and thus this is a loss function that includes a distribution loss term that regularizes an output of the model to produce realistic scenarios [0032]. Ambrus teaches generating a simulated scenario, with the simulated scenario including a position and pose (estimate position and orientation of the vehicle 100, [0065]) in a three-dimensional (3D) scene that is generated by the model [0003, 0027]; and training a self-driving model for an autonomous vehicle using the simulated scenario (vehicle 100 implementing the prediction system 170 may find the depth associated with other vehicles more relevant than the road for automated driving, as such, the prediction system 170 may perform training on a subset of estimated depth points associated with pixels using masking, [0026], vehicle 100 is an autonomous vehicle, autonomous vehicle refers to a vehicle that is capable of operating in an autonomous mode, autonomous mode refers to maneuvering the vehicle 100 using computing systems to control the vehicle 100 with no input from a human driver, [0042]).
However, Ambrus does not teach wherein the distribution loss term is configured to guide the model to increase sample density around a target region. However, Ha teaches wherein the distribution loss term is configured to guide the model to increase sample density around a target region (determining some points from among a plurality of points in the 3D scene as sample points in the 3D scene for rendering the plurality of points of the 3D scene such that a density of the extracted sample points is greater, adjusting a parameter of the sample point determination model to reduce a loss, where the loss is determined based on the density of the extracted sample points in the 3D scene, col. 20, lines 14-55).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Ambrus so that the distribution loss term is configured to guide the model to increase sample density around a target region because Ha suggests that this is needed for a greater diffuse reflection to produce a realistic 3D scene (col. 20, lines 14-55).
However, Ambrus and Ha do not teach generating the simulated scenario based on an original scenario, with the simulated scenario including a different position and pose relative to the original scenario in the 3D scene that is generated by the model from the original scenario. However, Ivanovic teaches generating a simulated scenario based on an original scenario, with the simulated scenario including a different position and pose relative to the original scenario in a three-dimensional (3D) scene that is generated by the model from the original scenario (input data 106 may include history data 110, [0042], history data 110 may include, but is not limited to, locations of the objects at time steps, and/or any other information, [0043], inputting the input data 106 and the goals data 114 into a conditional component 116 that is trained to generate data 118 representing actions associated with the navigational goals of the objects, for an object, the conditional component 116 may determine predicted trajectories associated with the object, conditional component 116 may predict the controls of the object, wherein the controls may include the respective location, and the like associated with the object, the conditional component 116 may then forward integrate the controls through the object’s dynamic model, [0055], simulation component 102 may receive the input data 106, wherein the input data 106 includes the prior locations of the objects, [0094], spatial component 112 may then process the input data 106 using the first machine learning model(s) and, based on the processing, output the goals data 114, the goals data 114 may represent the poses (e.g. locations, etc.) associated with the first object at a future time, [0095], 3D rendering, [0155]); and training a self-driving model for an autonomous vehicle using the simulated scenario (autonomous vehicle 1400, [0028], model(s) that is trained to generate data representing predicted trajectories of the objects, [0032], once the machine learning models are trained, the machine learning models may be used by the vehicles, [0202]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Ambrus and Ha to include generating the simulated scenario based on an original scenario, with the simulated scenario including a different position and pose relative to the original scenario in the 3D scene that is generated by the model from the original scenario because Ivanovic suggests that this allows for accurately simulating the trajectories of objects within the environment, even when performing simulations in new environments [0007].
As per Claim 3, Ambrus and Ha do not teach wherein generating the simulated scenario includes applying a criterion to limit a distance between the position of the simulated scenario and a position of the original scenario. However, Ivanovic teaches wherein generating the simulated scenario includes applying a criterion to limit a distance between the position of the simulated scenario and a position of the original scenario (continue to perform these processes until the simulated trajectories reach a threshold, such as a threshold distance (e.g., a threshold distance from the original locations of the objects at time 0 seconds), [0036]). This would be obvious for the reasons given in the rejection for Claim 1.
As per Claim 8, Ambrus and Ha do not teach further comprising sensing information about a new scenario and performing a driving action based on an output of the self-driving model responsive to the scenario. However, Ivanovic teaches further comprising sensing information about a new scenario and performing a driving action based on an output of the self-driving model responsive to the scenario (detect an impending forward collision with another vehicle or other object, and may automatically apply the brakes, [0187]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Ambrus and Ha to include sensing information about a new scenario and performing a driving action based on an output of the self-driving model responsive to the scenario because Ivanovic suggests that this way, if the driver does not take corrective action within a specified time or distance parameter, then the brakes are automatically applied in order to avoid collision with another vehicle or other object [0187].
14. As per Claim 9, Ambrus and Ha do not teach wherein the information about the new scenario includes new video and LiDAR information. However, Ivanovic teaches wherein the information about the new scenario includes new video and LiDAR information (front-facing cameras may be used to perform the same ADAS functions as LIDAR, including emergency braking, pedestrian detection, and collision avoidance, [0111], [0187]). This would be obvious for the reasons given in the rejection for Claim 8.
15. As per Claim 10, Ambrus and Ha do not teach wherein the driving action is selected from the group consisting of a steering action, an acceleration action, and a braking action. However, Ivanovic teaches wherein the driving action is selected from the group consisting of a steering action, an acceleration action, and a braking action [0187]. This would be obvious for the reasons given in the rejection for Claim 8.
16. As per Claim 11, Claim 11 is similar in scope to Claim 1, except that Claim 11 is directed to a system, comprising: a hardware processor; and a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to perform the method of Claim 1. Ambrus teaches a system, comprising: a hardware processor; and a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to perform the method (processes can be embedded in a computer-readable storage readable by a machine, tangibly embodying a program of instructions executable by the machine to perform methods described herein, [0070]). Thus, Claim 11 is rejected under the same rationale as Claim 1.
17. As per Claims 13 and 18-20, these claims are similar in scope to Claims 3, 8, 10, and 11 respectively, and therefore are rejected under the same rationale.
18. Claim(s) 2 and 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ambrus (US 20220301203A1), Ha (US012198245B2), and Ivanovic (US 20230391365A1) in view of Nygaard (US 20200050536A1).
19. As per Claim 2, Ambrus, Ha, and Ivanovic are relied upon for the teachings as discussed above relative to Claim 1.
However, Ambrus, Ha, and Ivanovic do not teach generating the simulated scenario includes applying a criterion to limit a distance between relative view direction and distance of simulated scenario and a relative view direction and distance of the original scenario. However, Nygaard teaches generating the simulated scenario includes applying a criterion to limit a distance between a relative view direction and distance of the simulated scenario and a relative view direction and distance of the original scenario (when the collision is determined to have occurred, flagging the simulation for further review, determining the divergence point includes determining when a location of the first simulated vehicle and a location of the vehicle from the log data diverge more than a threshold amount in a lateral direction relative to a direction of traffic in a lane in which the simulated vehicle is traveling in the first simulation, [0003]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Ambrus, Ha, and Ivanovic so generating the simulated scenario includes applying a criterion to limit a distance between relative view direction and distance of simulated scenario and a relative view direction and distance of the original scenario because Nygaard suggests this way, software can be rigorously tested without requiring a vehicle to physically drive real miles or having to manufacture situations in the real world [0018].
20. As per Claim 12, Claim 12 is similar in scope to Claim 2, and therefore is rejected under the same rationale.
21. Claim(s) 4 and 14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ambrus (US 20220301203A1), Ha (US012198245B2), and Ivanovic (US 20230391365A1) in view of Davison (US 20250068175A1).
22. As per Claim 4, Ambrus, Ha, and Ivanovic are relied upon for the teachings as discussed above relative to Claim 1.
However, Ambrus, Ha, and Ivanovic do not teach wherein the depth loss term is Ldepth (ϴ) = Er~D[(ẑ - z)2] where Er~D is an expected value for a ray sampled from ray distribution D, ẑ is an expected depth, and z is a depth from a training sample. However, Davison teaches wherein the depth loss term is Ldepth (ϴ) = Er~D[(ẑ - z)2] where Er~D is an expected value for a ray sampled from ray distribution D, ẑ is an expected depth, and z is a depth from a training sample [0056].
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Ambrus, Ha, and Ivanovic so that the depth loss term is Ldepth (ϴ) = Er~D[(ẑ - z)2] where Er~D is an expected value for a ray sampled from ray distribution D, ẑ is an expected depth, and z is a depth from a training sample because Davison suggests that this allows for an accurate scene representation to be obtained autonomously [0009].
23. As per Claim 14, Claim 14 is similar in scope to Claim 4, and therefore is rejected under the same rationale.
24. Claim(s) 6 and 16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ambrus (US 20220301203A1), Ha (US012198245B2), and Ivanovic (US 20230391365A1) in view of Irshad (US 20240171724A1).
25. As per Claim 6, Ambrus, Ha, and Ivanovic are relied upon for the teachings as discussed above relative to Claim 1.
However, Ambrus, Ha, and Ivanovic do not teach wherein the loss function has an RGB (red-green-blue) term implemented as a mean-squared error loss between a rendered image and a training sample. However, Irshad teaches wherein the loss function has an RGB (red-green-blue) term implemented as a mean-squared error loss between a rendered image and a training sample (for the first training phase, a mean squared error loss is employed on predicted color and target pixels at the sampled point locations in the ground-truth images, [0104]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Ambrus, Ha, and Ivanovic so that the loss function has an RGB (red-green-blue) term implemented as a mean-squared error loss between a rendered image and a training sample as suggested by Irshad. It is well-known in the art that mean squared error helps assess the accuracy of predictive models.
26. As per Claim 16, Claim 16 is similar in scope to Claim 6, and therefore is rejected under the same rationale.
27. Claim(s) 7 and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ambrus (US 20220301203A1), Ha (US012198245B2), and Ivanovic (US 20230391365A1) in view of Guha (US 20230350977A1).
28. As per Claim 7, Ambrus, Ha, and Ivanovic are relied upon for the teachings as discussed above relative to Claim 1.
However, Ambrus, Ha, and Ivanovic do not expressly teach training the self-driving model includes imitation learning of policy using simulated scenario and original scenario as examples. However, Guha teaches training the self-driving model includes imitation learning of policy using simulated scenario and original scenario as examples (machine learning model is trained using training data set that includes both original training data and simulated training data, [0014], synthetic photographs may be generated to depict various road conditions that may occur and may not be represented in training data used to train machine learning models that operate an autonomous vehicle, synthetic photographs may then be used to test existing machine learning models in order to discover further situations in which machine learning models may generate undesirable outputs, all synthetic photographs that are generated may be used to train, retrain, and test machine learning models for operating autonomous vehicles, [0062]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Ambrus, Ha, and Ivanovic so that training the self-driving model includes imitation learning of a policy using the simulated scenario and the original scenario as examples because Guha suggests that this way, the machine learning model is trained on further simulated scenarios, which helps to mitigate hurting or even killing pedestrians [0061-0062].
29. As per Claim 17, Claim 17 is similar in scope to Claim 7, and therefore is rejected under the same rationale.
Allowable Subject Matter
30. Claims 5 and 15 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
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 JONI HSU whose telephone number is (571)272-7785. The examiner can normally be reached M-F 10am-6:30pm.
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, Kee Tung can be reached at (571)272-7794. 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.
JH
/JONI HSU/Primary Examiner, Art Unit 2611
Prosecution Insights