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 Amendment
The Amendment filed 6/29/2026 has been entered. Claims 1-8, 10-11, and 13-14 have been amended. Claims 9, 12, and 15 have been cancelled. Claims 1-8, 10-11, and 13-14 are pending in the application.
Specification
The disclosure is objected to because of the following informalities.
First, the substitute paragraphs filed 6/29/2026 for page 22, lines 16-24, for page 23, lines 5-9, and for page 23, lines 19-24 each refer to Fig. 2a ("As shown in Fig. 2a, the system 50 comprises a providing unit 51"; "According to the embodiments shown in Fig. 2a, the system further comprises at least one sensor 54"; and "According to the embodiments of Fig. 2a, the set of predefined augmentation techniques"), but the system 50 and the elements 51 through 56 are shown only in Fig. 3, Fig. 2a shows only the graph structure 10 with its nodes 11 and starting node 12, and the unamended sentence preceding the first of those paragraphs states that "Fig. 3 illustrates a system for generating additional training data for training a machine learning algorithm 50". In each of the three paragraphs, "Fig. 2a" should read --Fig. 3--.
Second, pages 23-24 twice refers to "the shown system 10" ("The shown system 10 further comprises a data storage 56" and "Further, the shown system 10 is configured to execute a method"), whereas reference character 10 designates the graph structure of Fig. 2a and the system of Fig. 3 is designated 50.
Appropriate correction is required. No amendment to the drawings is required.
Claim Objections
Claim 1 is objected to because of the following informalities:
Claim 1, in the generating step, "modifying the position the at least one sensor" should read --modifying the position of the at least one sensor--.
Appropriate correction is required.
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.
Claims 1-4, 8, 10-11, and 13-14 are rejected under 35 U.S.C. 103 as being unpatentable over Bansal et al. (hereinafter Bansal) "ChauffeurNet: Learning to Drive by Imitating the Best and Synthesizing the Worst" (2018) in view of Yu et al. (hereinafter Yu) "Scene-Graph Augmented Data-Driven Risk Assessment of Autonomous Vehicle Decisions" (2021) and further in view of Ost et al. (hereinafter Ost) "Neural Scene Graphs for Dynamic Scenes" (2021).
Regarding independent claim 1, Bansal teaches a method for training a machine learning algorithm, the method comprising (Bansal: page 1, Abstract, "Our goal is to train a policy for autonomous driving via imitation learning"; Bansal: page 2, Section 1, "We present this mid-level input to a recurrent neural network (RNN), named ChauffeurNet"; Bansal trains the recurrent neural network ChauffeurNet (a machine learning algorithm) as a driving policy): providing training data for training the machine learning algorithm (Bansal: page 13, Section 6.1, "Our dataset contains approximately 26 million examples"; Bansal provides about 26 million driving examples (training data) for training ChauffeurNet), the training data including labeled sensor data from at least one sensor (Bansal: page 5, Section 3.1, "create it from real-sensor logs using a standard perception system that detects and tracks objects"; Bansal: page 13, Section 6.1, "we employ a separate perception system based on laser and camera data"; Bansal: page 10, Section 5.2.1, "the ground-truth boxes of all the scene objects at each timestep"; Bansal's examples are created from real-sensor logs of laser and camera data (at least one sensor), with ground-truth boxes of the detected scene objects attached to each example (labeled sensor data)); transforming the training data for training the machine learning algorithm (Bansal: page 2, Section 1, "We use a perception system that processes raw sensor information and produces our input: a top-down representation of the environment and intended route, where objects such as vehicles are drawn as oriented 2D boxes"; Bansal converts the logged sensor information into a top-down representation of the agent's environment in which each object is drawn at its location); generating additional training data for training the machine learning algorithm (Bansal: page 1, Abstract, "We propose exposing the learner to synthesized data in the form of perturbations to the expert's driving"; Bansal: page 9, Figure 5, "The perturbed example created by perturbing the current agent location (red point) in the original example away from the lane center"; Bansal synthesizes perturbed examples (additional training data) by moving the agent's current location within a logged example); and training, based on the training data and the additional training data, the machine learning algorithm to control a controllable function of a vehicle (Bansal: page 10, Section 5.1, "In training, we give perturbed examples a weight of 1/10 relative to the real examples"; Bansal: page 2, Section 1, "outputs a driving trajectory that is consumed by a controller which translates it to steering and acceleration"; Bansal trains ChauffeurNet on the real examples (the training data) together with the perturbed examples (the additional training data), and the trained network's trajectory is translated into the steering and acceleration (a controllable function) of the vehicle).
Bansal does not expressly teach algorithm in a graph structure, the graph structure including (i) a starting node that represents a position of the at least one sensor with respect to objects represented in the labeled sensor data.
However, Yu teaches algorithm in a graph structure, the graph structure including (i) a starting node that represents a position of the at least one sensor with respect to objects represented in the labeled sensor data (Yu: page 2, Figure 1, "Then, we project each frame to its bird’s-eye view to better approximate the spatial relations between objects. Finally, we construct a scene-graph using the list of detected objects and their attributes"; Yu: page 4, Section III-A-3, "adopt a simple rule to determine the relations between the objects using their attributes (e.g., relative location to the ego car)"; Yu: page 4, Section III-A-3, "we use each vehicle’s horizontal displacement relative to the ego vehicle to assign vehicles to either the Left Lane, Middle Lane, or Right Lane"; Yu converts each camera frame of a driving clip (the labeled sensor data) into a scene-graph (algorithm in a graph structure) and builds its distance, direction and lane relations from each detected object's position relative to the ego car, so the Ego Car node (a starting node) is the node from which the scene-graph is built; and because projecting a single camera frame to a bird's-eye view yields each object's location relative to the ego vehicle only when the frame was captured from that vehicle, the Ego Car node represents the position of the camera (the at least one sensor) with respect to the detected objects (objects represented in the labeled sensor data)).
Because Bansal and Yu are analogous art with both addressing machine learning for autonomous-vehicle driving decisions from a mid-level representation of a sensed traffic scene, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Yu's scene-graph to the method of Bansal, with a reasonable expectation of success, by building for each example a scene-graph around an ego-vehicle node from the location and class of each object that Bansal's perception system detects, as Yu builds its scene-graphs from object location and class information (Yu: page 4, Section III-A-2, "we build our scene-graphs using the ground-truth location and class information for each vehicle"), and by encoding that scene-graph with Yu's graph network as an additional input to the driving network alongside the top-down images that Bansal continues to render from the same objects (Bansal: page 7, Section 3.3, "The environment information is rendered into the input images described in Fig. 1"), the scene-graph adding what Yu describes as an intermediate representation "that encodes the spatial and semantic relations between all the traffic participants in a frame" (Yu: page 3, Section II-A), to teach transforming the training data for training the machine learning algorithm in a graph structure, the graph structure including (i) a starting node that represents a position of the at least one sensor with respect to objects represented in the labeled sensor data. This modification would have been motivated by the desire to better transfer knowledge gained from simulated environments to real-world tasks (Yu: page 2, Section I).
Bansal and Yu do not expressly teach and (ii) nodes representing positions of the objects; training by modifying the graph structure, the modifying of (interpreted per the Claim Objections set forth above) the graph structure including at least one of (i) modifying the position the at least one sensor and (ii) modifying at least one of the positions of the objects.
However, Ost teaches and (ii) nodes representing positions of the objects (Ost: page 4, Section 3.2, "Each object is represented by a neural radiance field in the local space of its node and position po"; Ost: page 3, Section 3 Graph Definition, "For a given scene graph, poses and locations for all objects can be extracted"; Ost's scene graph of a recorded driving scene represents each dynamic object at its own node together with its position po, so its object nodes represent the positions of the objects); training by modifying the graph structure, the modifying the graph structure including at least one of (i) modifying the position the at least one sensor and (ii) modifying at least one of the positions of the objects (Ost: page 6, Figure 6, "A learned vehicle leaf node is translated by 2 meters perpendicular to the movement observed during training"; Ost: page 7, Figure 8, "While the ego camera is moved by approximately 2 m into the scene, all other nodes of the scene graph are fixed"; Ost modifies its learned scene graph by translating a vehicle node away from the movement observed during training (modifying at least one of the positions of the objects) and, separately, by moving the ego camera while every other node stays fixed (modifying the position the at least one sensor); the alternatives require only one member, and Ost teaches both).
Because Bansal, in view of Yu, and Ost are analogous art with all three addressing machine-learning representations of sensed traffic scenes for autonomous driving, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Ost's position-carrying object nodes and node manipulations to the method of Bansal, in view of Yu, with a reasonable expectation of success, storing each detected object's position at its node of the ego-anchored scene-graph and carrying out Bansal's perturbation of the current agent location (Bansal: page 9, Figure 5) on that scene-graph by moving the ego-vehicle node, and likewise moving or removing an object's node, the relations of the scene-graph then being re-derived from the changed positions by Yu's location-based rule (Yu: page 4, Section III-A-3, "we extract only the location information for each object and adopt a simple rule to determine the relations between the objects using their attributes"), thereby teaching transforming the training data for training the machine learning algorithm in a graph structure, the graph structure including (i) a starting node that represents a position of the at least one sensor with respect to objects represented in the labeled sensor data and (ii) nodes representing positions of the objects; generating additional training data for training the machine learning algorithm by modifying the graph structure, the modifying of the graph structure including at least one of (i) modifying the position the at least one sensor and (ii) modifying at least one of the positions of the objects. This modification would have been motivated by the desire to modify learned scene graphs to synthesize unseen frames of novel object arrangements (Ost: page 6, Section 5).
Regarding dependent claim 2, Bansal, in view of Yu and Ost, teach the method of claim 1, wherein the providing the training data comprises: acquiring sensor data with the at least one sensor (Bansal: page 13, Section 6.1, "we employ a separate perception system based on laser and camera data"; Bansal acquires laser and camera data (sensor data) with the vehicle's laser and camera (the at least one sensor)); and labeling the acquired sensor data based on the objects represented in the acquired sensor data to generate the training data (Bansal: page 5, Section 3.1, "create it from real-sensor logs using a standard perception system that detects and tracks objects"; Bansal's perception system detects and tracks the objects in the real-sensor logs and the training examples are created from those detections, so the acquired sensor data is labeled based on the objects it represents to generate the training data).
Regarding dependent claim 3, Bansal, in view of Yu and Ost, teach the method of claim 1, wherein the generating the additional training data comprises: dropping at least one node of the graph structure (Ost: page 6, Section 5.1, "A learned neural scene graph can be manipulated at its edges and nodes"; Ost: page 7, Section 5.1, "In addition to pose manipulation and node removal from a learned scene graph, our method allows for constructing completely novel scene graphs"; Ost lists node removal, alongside pose manipulation, among the manipulations of a learned scene graph (the graph structure) of a recorded driving scene from which Ost renders scenes that were not observed, so removing an object node (at least one node) from that graph is dropping at least one node of the graph).
Regarding dependent claim 4, Bansal, in view of Yu and Ost, teach the method of claim 1, wherein the generating the additional training data comprises: adding a perturbation to at least one node of the graph structure (Ost: page 6, Figure 6, "A learned vehicle leaf node is translated by 2 meters perpendicular to the movement observed during training"; translating a leaf node by a fixed offset is a perturbation applied to a node of the graph).
Regarding dependent claim 8, Bansal, in view of Yu and Ost, teach the method of claim 1, wherein the generating the additional training data comprises: moving the starting node of the graph structure (Ost: page 7, Figure 8, "While the ego camera is moved by approximately 2 m into the scene, all other nodes of the scene graph are fixed"; Ost: page 7, Section 5.1, "Fig. 8 reports novel views for ego-vehicle motion into the scene, where the ego-vehicle is driving ≈2 m forward"; Ost moves the ego camera (the at least one sensor) forward while every other node of the scene graph stays fixed, so the position of the ego camera relative to every other node changes, in the combination set forth above for claim 1, the node so moved is the ego-vehicle node from which the scene-graph is built, that is, the starting node).
Regarding claim 10, Bansal, in view of Yu and Ost, teach a method for controlling the controllable function of the vehicle based on the machine learning algorithm, comprising: providing the machine learning algorithm for controlling the controllable function, wherein the machine learning algorithm is trained by the method according to claim 1; and controlling the controllable function based on the trained machine learning algorithm (Bansal: page 7, Section 3.3, "given to the RNN which then outputs a future trajectory. This is fed to a controls optimizer that outputs the low-level control signals that drive the vehicle"; Bansal's self-driving system provides the trained recurrent neural network ChauffeurNet (the machine learning algorithm) and drives the vehicle with control signals derived from its output (controlling the controllable function)). The substantive limitations of claim 10 that incorporate claim 1 are taught by Bansal, in view of Yu and Ost, for the same reasons set forth above for claim 1. The motivation to combine is the same as set forth above for claim 1.
Regarding claim 11, Bansal, in view of Yu and Ost, teach a system for training the machine learning algorithm, the system comprising: a processor configured to execute the method according to claim 1 (Yu: page 6, Section IV-B, "All the experiments were conducted on a server with one NVIDIA TITAN-XP graphics card and one NVIDIA GeForce GTX 1080 graphics card"; Yu trains and evaluates its models on a server with graphics cards (a processor)). The substantive limitations of claim 11 that incorporate claim 1 are taught by Bansal, in view of Yu and Ost, for the same reasons set forth above for claim 1. The motivation to combine is the same as set forth above for claim 1.
Regarding claim 13, Bansal, in view of Yu and Ost, teach a system for controlling the controllable function of the vehicle based on the machine learning algorithm, the system comprising: a non-transitory memory configured to store the machine learning algorithm for controlling the controllable function, wherein the machine learning algorithm is trained by the method according to claim 1, (Bansal: page 13, Section 6.1, "The model runs on a NVidia Tesla P100 GPU in 160ms"; a GPU that runs the model necessarily holds the model's parameters in its memory (a non-transitory memory) while running it) and a controller configured to control the controllable function based on the machine learning algorithm (Bansal: page 2, Section 1, "outputs a driving trajectory that is consumed by a controller which translates it to steering and acceleration"; Bansal's controller (a controller) translates the network's trajectory into steering and acceleration). The substantive limitations of claim 13 that incorporate claim 1 are taught by Bansal, in view of Yu and Ost, for the same reasons set forth above for claim 1. The motivation to combine is the same as set forth above for claim 1.
Regarding claim 14, Bansal, in view of Yu and Ost, teach a non-transitory computer-readable medium that stores a computer program having instructions that, when executed by a computer, cause the computer to perform the method according to claim 1 (Yu: page 6, Section IV-B, "Our models were implemented using PyTorch and PyTorch-Geometric"; Yu: page 6, Section IV-B, "All the experiments were conducted on a server with one NVIDIA TITAN-XP graphics card and one NVIDIA GeForce GTX 1080 graphics card"; Yu implements its models as programs (a computer program having instructions) that a server (a computer) executes, and a program a server executes is necessarily stored on that server's storage (a non-transitory computer-readable medium)). The substantive limitations of claim 14 that incorporate claim 1 are taught by Bansal, in view of Yu and Ost, for the same reasons set forth above for claim 1. The motivation to combine is the same as set forth above for claim 1.
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Bansal in view of Yu and Ost and further in view of Zhang et al. (hereinafter Zhang), US 2021/0073660 A1.
Zhang was disclosed in an IDS dated 9/18/2023
Regarding dependent claim 5, Bansal, in view of Yu and Ost, teach the method of claim 1, wherein the generating the additional training data comprises: at least a part of the graph structure (Ost: page 4, Section 3.2, "Each object is represented by a neural radiance field in the local space of its node and position po"; Ost: page 6, Section 5.1, "A learned neural scene graph can be manipulated at its edges and nodes"; Ost manipulates selected nodes of a learned scene graph, each dynamic object node carrying its object's position po, so the portion of the scene graph that Ost manipulates is at least a part of the graph structure).
Bansal, Yu, and Ost do not expressly teach mirroring at least a part of the graph structure.
However, Zhang teaches mirroring (Zhang: [0032], "the different data augmentation techniques may comprise rotation (g0), mirroring (g1), adding noise (g2)"; Zhang: [0032], "s1 determining horizontal or vertical mirroring"; Zhang derives a new data instance for training a machine learnable model from an existing data instance by applying a mirroring augmentation (mirroring) whose parameter selects horizontal or vertical mirroring).
Because Bansal, in view of Yu and Ost, and Zhang are analogous art with all four addressing the generation of additional data from recorded sensor data for training machine learning models used with vehicles, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Zhang's horizontal mirroring to the method of Bansal, in view of Yu and Ost, with a reasonable expectation of success, Zhang listing mirroring beside rotation (Zhang: [0032]), the rotation that Bansal already applies to its top-down driving view for data augmentation (Bansal: page 3, Section 3.1, "For data augmentation purposes during training, the orientation of the coordinate system is randomly picked for each training example"), by reversing the sign of the lateral coordinate of the position stored at each object node of the scene graph, the graph counterpart of mirroring horizontally the top-down view that Bansal renders from the same detected objects, so that the positions carried by the object nodes are reflected across the ego vehicle's direction of travel, with the recorded trajectory target reflected in the same way, thereby teaching wherein the generating the additional training data comprises: mirroring at least a part of the graph structure. This modification would have been motivated by the desire to prevent or reduce overfitting of the machine learnable model to the training data (Zhang: [0011]).
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Bansal in view of Yu and Ost and further in view of You et al. (hereinafter You) "Graph Contrastive Learning with Augmentations" (2020).
Regarding dependent claim 6, Bansal, in view of Yu and Ost, teach the method of claim 1, wherein the generating the additional training data comprises: at least one feature of at least one object represented in the labeled sensor data (Yu: page 3, Section III-A-1, "we use its estimated location and class type"; Yu: page 5, Section III-B-1, "by directly converting the node’s type information to its corresponding one-hot vector"; for each object detected in a camera frame Yu computes the object's estimated location and class type and encodes the class type as the feature vector of that object's node, so each node carries at least one feature of at least one object represented in the labeled sensor data).
Bansal, Yu, and Ost do not expressly teach masking at least one feature of at least one object represented in the labeled sensor data.
However, You teaches masking (You: page 3, Section 3.1, "Attribute masking prompts models to recover masked vertex attributes using their context information, i.e., the remaining attributes"; You: page 3, Section 3.1, "Data augmentation aims at creating novel and realistically rational data through applying certain transformation without affecting the semantics label"; You creates augmented graphs from an input graph, for learning graph neural network representations, by masking part of the attributes of its vertices (masking) while the remaining attributes serve as context).
Because Bansal, in view of Yu and Ost, and You are analogous art with all four addressing the generation of augmented training data for learning models that operate on graph or other intermediate representations of their input, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply You's attribute masking to the method of Bansal, in view of Yu and Ost, with a reasonable expectation of success, by setting to zero the class-type entries of the feature vector of a detected object's node in a copy of the scene graph while that node keeps its position and relations as the remaining context, just as Bansal already blanks part of a driving example's input during training (Bansal: page 8, Section 4.2, "for 50% of the examples, we keep only the current position (u0, v0) of the agent in the past agent poses channel of the input data"), thereby teaching wherein the generating the additional training data comprises: masking at least one feature of at least one object represented in the labeled sensor data. This modification would have been motivated by the desire to learn graph representations of similar or better generalizability, transferrability, and robustness (You: page 1, Abstract).
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Bansal in view of Yu and Ost and further in view of Shawn Hunt (hereinafter Hunt), US 2021/0406597 A1.
Regarding dependent claim 7, Bansal, in view of Yu and Ost, teach the method of claim 1, wherein the generating the additional training data comprises: at least one node in the graph structure (Ost: page 6, Section 5, "we optimize a neural scene graph with a node for each tracked object and one for the static background"; Ost represents each tracked object of a recorded driving scene by its own node (at least one node) of the scene graph (the graph structure)).
Bansal, Yu, and Ost do not expressly teach duplicating at least one node in the graph structure.
However, Hunt teaches duplicating (Hunt: [0045], "The duplication module 172 causes the processor(s) 110 to duplicate the original object 183B by duplicating the portion of the points 504B of the point cloud and the annotation data 506B"; Hunt: [0045], "this duplicated portion of the points 504B of the point cloud and the annotation data 506B can then be inserted into the training data 500 as a third object"; Hunt: [0046], "The synthetic object 187A is essentially a duplicate of the original object 183B but has been rotated"; Hunt copies the points and annotation data of an original object 183B of the training data 500 and inserts the copy into the same training data 500 as a third object, so the duplication module 172 performs the duplicating and the original object 183B is then represented twice, a copy that has been rotated remaining a duplicate of that object).
Because Bansal, in view of Yu and Ost, and Hunt are analogous art with all four addressing the generation of additional training data from sensor data of a vehicle's surroundings for training machine learning models, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Hunt's duplication of an annotated object to the method of Bansal, in view of Yu and Ost, with a reasonable expectation of success, Hunt stating that its training data "could be a multidimensional data structure" (Hunt: [0040]), by copying an object node of the scene graph with its attributes and inserting the copy into the same scene graph as a further object node, placed with respect to the ego vehicle as Hunt places its copies (Hunt: [0049]) and at a transformation on the road lanes that avoids collisions as Ost places sampled nodes (Ost: page 7, Figure 7, "new transformations sampled on the road lanes, not allowing for collisions between objects"), and clear of the target geometry that Bansal renders from the recorded waypoints of each example (Bansal: page 10, Section 5.2.3, "We model this target geometry by fitting a smooth curve to the target waypoints"), so the recorded targets remain valid, while Bansal's collision loss takes the inserted copy into account from the objects the example contains (Bansal: page 10, Section 5.2.1, "a binary mask with ones at all pixels occupied by other dynamic objects") without a separate annotation, thereby teaching wherein the generating the additional training data comprises: duplicating at least one node in the graph structure. This modification would have been motivated by the desire to add already annotated objects to training data quickly and without manual annotation, making for more complex training data (Hunt: [0053]).
Response to Arguments
Applicant's arguments in the Remarks filed 6/29/2026 have been fully considered and responded to below.
Regarding the objections to the drawings and to the specification, on page 11 of the Remarks, Applicant states that pages 22-23 of the specification refer to the incorrect figure in several instances and that "The specification has been amended as required". The argument is persuasive as to the drawings and as to the substitute paragraphs for page 22, lines 2-4 and lines 9-11, which now refer to Fig. 2c and Fig. 2d, the figures that show the modified graph structures 30 and 40. The drawings as filed show that subject matter, so the objection to the drawings under 37 CFR 1.83(a) is withdrawn and no corrected drawing sheet is required. The argument is not persuasive as to the substitute paragraphs for page 22, lines 16-24, page 23, lines 5-9, and page 23, lines 19-24, which replace one incorrect figure reference with another, and the objection to the specification is maintained to that extent, as set out above.
Regarding the rejection under 35 U.S.C. 112(b), on page 11 of the Remarks, Applicant states that "the claims are amended to clarify the issue noted by the Office". The argument is persuasive. Claims 9 and 12 are canceled. Claim 10 now begins "A method for controlling the controllable function of the vehicle", so the definite article that lacked antecedent basis is gone, and claim 13 no longer recites a control unit that comprises a control unit, the double inclusion on which the rejection rested. The rejection of claims 10 and 13 under 35 U.S.C. 112(b) is withdrawn, and the rejection is moot as to claims 9 and 12.
Regarding the rejection of claims 11, 14 and 15 under 35 U.S.C. 101 as directed to non-statutory subject matter, on page 11 of the Remarks, Applicant states that claim 11 "is amended to recite the structural component of" a processor and that claim 14 is amended to recite a non-transitory computer-readable medium that stores a computer program. The argument is persuasive. Amended claim 11 positively requires a processor, and amended claim 14 is directed to a non-transitory computer-readable medium, so the broadest reasonable interpretation of neither claim reaches software per se or a transitory signal (MPEP 2106.03). The rejection of claims 11 and 14 is withdrawn, and the rejection is moot as to canceled claim 15.
Regarding the rejection under 35 U.S.C. 101, Step 2A Prong One, on pages 12-13 of the Remarks, Applicant argues that "it is clearly not practical to train a machine learning algorithm using the human mind or with the aid of pen and paper". The argument is persuasive as to the training limitation, which the prior action did not treat as part of the judicial exception, but it does not reach the transforming and generating steps, which as recited can still be performed in the mind; a claim recites a judicial exception even when it also recites limitations that are not themselves an exception (MPEP 2106.04(a)). The eligibility of the claims is resolved instead at Step 2A Prong Two, as set out next.
Regarding Step 2A Prong Two, on pages 13-16 of the Remarks, Applicant argues that the claimed invention "transforms the provided training data into a graph structure in which a starting node represents the position of the at least one sensor with respect to objects represented in the sensor data" and trains the machine learning algorithm to control a controllable function of a vehicle. The argument is persuasive. Amended claim 1 recites the mechanism by which the asserted improvement is obtained, namely generating additional training data by modifying the positions represented in a graph structure whose starting node represents the position of the sensor, and training the machine learning algorithm on the training data and the additional training data to control a controllable function of a vehicle, rather than reciting the result alone. The additional elements therefore integrate the recited judicial exception into a practical application (MPEP 2106.04(d); MPEP 2106.05(a)). The rejection of claims 1-8, 10, 11, 13 and 14 under 35 U.S.C. 101 is withdrawn, and it is moot as to canceled claims 9, 12 and 15.
Regarding Step 2B, on page 16, Applicant argues that the claims "recite additional limitations which amount to significantly more than the abstract idea". Because the rejection is withdrawn at Step 2A Prong Two, Step 2B is not reached, and the argument is moot.
Regarding the rejection of claim 1 under 35 U.S.C. 103 over Yu in view of You, on pages 16-19, Applicant argues that "You fails to teach modifying the graph structure by modifying the position the at least one sensor or modifying at least one of the positions of the objects" because "the node dropping, edge perturbation, attribute masking, and subgraph techniques for graph augmentation do not constitute modifying a position of a sensor or of an object in the graph structure." The argument is persuasive as to the ground of record, and the rejection of claims 1-4, 6, 11 and 14 over Yu in view of You is withdrawn. However, the argument does not apply to the new ground of rejection of claim 1, which does not rely on You for any limitation of claim 1. That ground relies on Ost for nodes representing positions of the objects and for modifying the position of the sensor and the positions of the objects, Ost translating a vehicle node and moving the ego camera of its scene graph (Ost: page 6, Figure 6; page 7, Figure 8), and relies on Bansal for generating additional training data and for training the machine learning algorithm on the training data and the additional training data to control a controllable function of a vehicle. Yu is relied upon only for the graph structure and its starting node. You is applied only to claim 6, for masking, which the argument does not address.
Regarding claims 2-4, 6, 9, 11, 12, 14 and 15, which Applicant argues only through their dependence from claim 1, on page 19 of the Remarks, the argument is persuasive as to the ground of record, which is withdrawn, and it is moot as to canceled claims 9, 12 and 15. Claims 2-4, 11 and 14 are rejected on new grounds over Bansal in view of Yu and further in view of Ost, and claim 6 over those references and further in view of You, as set out above. The argument advances no reason specific to any of those claims.
Regarding claims 5, 7, 8, 10 and 13, on page 19 of the Remarks, as to which Applicant argues that "Zhang also does not teach these limitations". The rejection of those claims over Yu in view of You and further in view of Zhang is withdrawn in view of the amendment, and new grounds of rejection are stated above. In those grounds Zhang is relied upon only for the mirroring of claim 5 (Zhang: [0032]). Claims 8, 10 and 13 are rejected over Bansal in view of Yu and further in view of Ost, and claim 7 over those references and further in view of Hunt. The argument identifies no teaching for which Zhang is relied upon that Zhang lacks, and one cannot show nonobviousness by attacking references individually where the rejection is based on a combination of references (MPEP 2145, subsection IV).
In summary, Applicant's arguments directed to the objection to the drawings, to the rejection under 35 U.S.C. 112(b), to the rejections under 35 U.S.C. 101, and to the rejections under 35 U.S.C. 103 of record are persuasive, and those grounds are withdrawn. Applicant's argument directed to the objection to the specification is persuasive only in part, and that objection is maintained as to the three substitute paragraphs identified above. Applicant's arguments do not apply to the new grounds of rejection under 35 U.S.C. 103 set forth in this action.
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 KUANG FU CHEN whose telephone number is (571)272-1393. The examiner can normally be reached M-F 9:00-5:30pm ET.
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, Jennifer Welch can be reached on (571) 272-7212. 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.
/KC CHEN/Primary Patent Examiner, Art Unit 2143
Prosecution Insights